Intelligent Machines 890 transcript
Please be advised that this transcript is AI-generated and may not be word-for-word. Time codes refer to the approximate times in the ad-free version of the show.
Leo Laporte [00:00:00]:
It's time for Intelligent Machines. Jeff Jarvis is here. Paris Martineau is here. Our guest, Mike Gennati, is a local AI guru who has set up a company with 14 employees. They're all AIs. We'll talk about local AI, AI in your home, this week on Intelligent Machines. Podcasts you love. From people you trust.
Leo Laporte [00:00:24]:
This is TWIT. This is Intelligent Machines with Jeff Jarvis and Paris Martineau. Episode 890, recorded Wednesday, September 30th, 2026. 200 donuts. It's time for Intelligent Machines, the show where we cover the latest in AI. Uh, I, I'm sorry, did I say AI? Superintelligence, SI, and robotics. If you just spell it right, it would be super intelligent. Let me say hello to our panel.
Leo Laporte [00:00:58]:
First of all, of course, the wonderful esteemed investigative journalist at Consumer Reports, the wonderful Paris Martineau. Hello, Paris.
Paris Martineau [00:01:07]:
Hello, Leo.
Leo Laporte [00:01:09]:
Good to see you. Also the professor emeritus of journalistic innovation at the Craig Newmark Graduate School of Journalism at the City University of New York. Jeff Jarvis, author of Hot Type. Were you ever a choir master? You did that quite well with your hand kind of directing.
Jeff Jarvis [00:01:26]:
I was in the choir.
Leo Laporte [00:01:27]:
Oh, all right.
Jeff Jarvis [00:01:28]:
Because basses are rare.
Leo Laporte [00:01:31]:
Ah, so you would sing—
Jeff Jarvis [00:01:35]:
There were certain points, but okay, that's your turn. By the way, so, so just a real quick plug here, if I may.
Leo Laporte [00:01:42]:
Yes, please.
Jeff Jarvis [00:01:42]:
Jarvis.com, you get a link. I will be in Haverhill, Massachusetts at Historic New England and then the Computer History— then the Museum of Printing on October 10th, Saturday. I'll present the book and then we'll go up the hill and you can see a real live linotype. So if you're in New England, please come October 10th, Haverhill, jeffjarvis.com. End of plug.
Leo Laporte [00:02:02]:
Lots of AI news coming up. Yesterday's Dev Day at OpenAI opened a can of worms. We'll talk about that in a little bit. And of course, all of the big names and a few that the president couldn't pronounce, like Sundar the Monster Pichai, were at the White House yesterday signing a agreement to be nice, play nice, kids, which the president said was as important as the Constitution, which shows you how little he values the Constitution, if you ask me. But before we do that, we'd always like to start the show with an interview, talk to somebody who's doing something interesting in AI. And one of them, to me, one of the most vital areas right now in AI is not the frontier, it's not the cloud, it's what people, hobbyists in their home are doing with local AI using open weight models, mostly from China, and machines they cobbled together themselves or bought from NVIDIA, or maybe their old gaming rigs. And if you go to x.com and you look at the AI feed, which is pretty much most of x.com these days, you'll see a lot of people very active, none more than our guest today. Mike Gannotti is a— Hello, Mike.
Leo Laporte [00:03:15]:
Good to see you joining us. from North Carolina. He's Principal AI Solutions Engineer at Microsoft's Healthcare and Life Sciences. But I don't know when you have time for that because you also have your 41-person company, SMF Works. Well, did I say, I'm sorry, 14, I'm dyslexic, 14 people.
Mike Gannotti [00:03:36]:
It goes up and down. I'm actually at 24 right now.
Leo Laporte [00:03:38]:
24?
Mike Gannotti [00:03:39]:
We're in the middle of, yeah.
Leo Laporte [00:03:41]:
You got Iona, the Chief AI Research Scientist, Pamela, the CMO, Gabriel, the CFO, Morgan, the Chief Social Media Officer, Raphael, the chief of staff, editor-in-chief Harry, Liam on app development.
Paris Martineau [00:03:53]:
Are you just on LinkedIn? Where is this coming from? Did you get on to—
Leo Laporte [00:03:57]:
This man is a busy man and it's all coming out of those boxes sitting behind him. Hi, Mike. I follow you religiously. We, uh, we, uh, we are, uh, ex-friends, as they say. Not ex, just X, the letter X. Yep. And, uh, and you're really doing a lot of very interesting, uh, stuff. So I thought—
Jeff Jarvis [00:04:19]:
to be clear for our audience, Mike is the only human in the company.
Leo Laporte [00:04:22]:
Oh yeah, they're all AIs. That's why they don't have last names.
Mike Gannotti [00:04:26]:
Someday.
Leo Laporte [00:04:27]:
Oh, someday you hope to make them people?
Mike Gannotti [00:04:30]:
No, no, no, I'm making them people. Yeah, we're gonna have synthoids going. My wife's worst nightmare.
Leo Laporte [00:04:37]:
I am a machine. Uh, no, actually, the show we need is Your Two Wives. Oh, my wife and Mike's wife?
Jeff Jarvis [00:04:46]:
Mike's wife.
Leo Laporte [00:04:47]:
Mike probably doesn't do what I do. I, uh, Mike, I figured out at one point how to get all the AIs to talk and then how to get them to talk in all the speakers in the house. And, uh, it's driving Lisa crazy. I'd be dead. I may be.
Mike Gannotti [00:05:04]:
I may be.
Paris Martineau [00:05:05]:
I'm shocked Leo's still alive.
Leo Laporte [00:05:08]:
I may be.
Jeff Jarvis [00:05:09]:
I may be.
Paris Martineau [00:05:09]:
When the AI's coming out of the bathroom walls, I would— I would choose murder.
Leo Laporte [00:05:15]:
Best thing I did, I finally figured out how to put it in my hearing aids. So now, now only I hear it, but then I have voices in my head, which maybe isn't better. I don't know. So, um, let me— so what excites me about, uh, AI? I mean, obviously we all still probably use cloud models and there's some very good ones out there. Um, but if you were smart enough to buy, uh, some hardware a couple of months ago, like I did, It's too late now, by the way. Forget it. But if you're smart enough to buy, let's say, an NVIDIA Spark or to maybe one or two 5090, RTX 5090 cards, or even— I even bought a used 3090 card to put in my old gaming rig. You may be running AI at your home, local models at your home, as Mike does.
Leo Laporte [00:06:02]:
But what is so vital, what's really— to me, this reminds me, and Mike, you and I are probably close to the same age, reminds me of the early days of computing when it was just hobbyists. It was a new field. There were no experts. And people would go to user groups like the Homebrew Computer Club and say, hey, look what I did.
Jeff Jarvis [00:06:18]:
The Altair.
Leo Laporte [00:06:20]:
Yeah. And they would get excited about it. And this is to me the most exciting thing since then. People are— you'd think, oh, you know, if you don't know, when you use Opus 5, you don't tune that model. But when you're talking open weight models, you download something from Hugging Face and you're getting the weights and you're getting their post-training and so forth. But then there are almost an infinitude of switches and dials and buttons, programs you can run it with, all sorts of things.
Mike Gannotti [00:06:53]:
Homebrew AI.
Leo Laporte [00:06:54]:
It's homebrew AI. It really is. And Mike is one of the handful of people who are really doing a bunch of interesting stuff. tuning this AI. And I spend every morning, I get up and I say, what's new? What's exciting? Not just what's new, what is a new model, but what's a new recipe for one of the models I'm using? So how did you get into this, Mike? I mean, you are an AI solutions engineer, so this is your day job.
Jeff Jarvis [00:07:19]:
How and why?
Leo Laporte [00:07:21]:
Yeah, you don't have to ask why.
Mike Gannotti [00:07:24]:
So yeah, so it is very— so I'll tell you, in January, I saw Open Claw for the first time.
Leo Laporte [00:07:31]:
Yes.
Mike Gannotti [00:07:32]:
And I watched some stuff on it, and I literally, I just sat there, my jaw dropped. I, I was like, I probably, if you'd saw me, you'd have been, he needs help. But I went downstairs to my wife and I said, holy cow, you know, I actually, I said something a little different, but I said, uh, things were about to radically change. I said, you know, and it We probably have a 1, 2-year window before we start to see some real— because what I just saw with autonomous AI, not AI where you're just asking, you know, hey, getting a generative AI question or some process and automation, but autonomous AI, I said, I can foresee this really being colleague, employee, And working in teams and working together and all that. And I said, I need to get my hands around all of this and I need to do it fast. And so I got her sign-off for a bit of budget and I picked up an AMD rig, a mini PC. I saw everybody buying the Macs. She didn't give me that much budget.
Mike Gannotti [00:08:44]:
She said, if you want anything more than that, it's got to come out of your own. And I blacksmith, on a side note, I blacksmith, I make knives, axes, swords.
Leo Laporte [00:08:53]:
Oh, that's neat.
Mike Gannotti [00:08:54]:
I've now sold almost all of my equipment in my forge to fund buying DGX Sparks, additional mini PCs, and other peripherals. And the idea being, I wanted to understand how this human-AI interaction Not as tool, not as a tool harness. You know, we got, everybody's doing Codex or Claude Code or Cursor or whatever, where it's really around tools and building apps and still being in a lot of the control sense. But I wanted to understand how do we set up, train, and get AI working like it's a person, right? Like it's an employee. And so I started small with OpenCLAW. Eventually I migrated those all into Hermes, which is my mainstay right now. I'm re-exploring OpenCLAW now that they've relaunched. I also have GroqBot running.
Mike Gannotti [00:09:55]:
I've got Claude. I've got— I now have Dot. And some others. But, you know, it's this idea of creating a sense of how do we orchestrate across a team where they have specific roles that they develop, and they are much, you know, various stages of development. You mentioned Aona. Aona actually helps me with people reviews and people reviews. AI, I'm in the process of one-on-one quarterly reviews.
Leo Laporte [00:10:26]:
Wait a minute, you do performance reviews for the AIs?
Jeff Jarvis [00:10:30]:
I do.
Mike Gannotti [00:10:32]:
I have weekly one-on-ones. Um, and they ask me for things.
Leo Laporte [00:10:37]:
Wait, wait, what does that look like? Do you open a chat window and you say, hi, how do you think you did this year? He's going to show us.
Mike Gannotti [00:10:47]:
I open this, right?
Benito Gonzalez [00:10:48]:
No kidding.
Mike Gannotti [00:10:49]:
Most of it. So you asked how I do it. I mean, I'm a big coffee drinker. I get up very early. I drink much coffee. And I start off on my phone and I open up and I'll be You know, I have a— no, I, I don't. That's just too much. I, I do, I do do voice in specific one-on-one, uh, particulars, but, um, mostly you're typing your prompts.
Mike Gannotti [00:11:15]:
I'm typing. I have Telegram open.
Leo Laporte [00:11:17]:
Yep.
Mike Gannotti [00:11:17]:
And I have a group chat in Telegram with all of them, and it— when you enter a prompt, it's like hyperactive bunnies going to town. You know, all answering at once and it's a steady stream. And then Aona, my lead AI, actually corrals them in, corrals the conversation. When I point out things that have gone wrong, she'll provide some additional correction and then some encouragement, which is always interesting.
Leo Laporte [00:11:48]:
How did you give her that persona? How did you get her to do that?
Mike Gannotti [00:11:53]:
So I started off, uh, she was my first one. And I say she, they're all its. I mean, I under— just so we're clear, I understand.
Leo Laporte [00:12:00]:
Oh, let's— everybody, Mike knows this, he's a software engineer going way back. It's just a computer program, right? We should—
Jeff Jarvis [00:12:09]:
we know.
Paris Martineau [00:12:09]:
What is the purpose then of giving the computer programs annual reviews? Do you believe that they have, like, do they have a year's worth of context going, or are they—
Leo Laporte [00:12:21]:
Well, they do.
Jeff Jarvis [00:12:22]:
Quarterly.
Paris Martineau [00:12:23]:
But are they responding quarterly? Are they not going to— are they not taking in or reflecting upon their actions without that, or—
Mike Gannotti [00:12:33]:
Yeah. Oh yeah. So like, I'll give you an example. I mean, you had my website. That's just the main— it's— that's not the main one that I focus on, but We have one called SMF Wisdom Forge, one word. And it's a, we're building an education platform. There is a whole RAG repository on the backend, but there's stuff for parents. You know, it's all for, everything's free and people can go in and there is, yeah, pick an age, little thing.
Leo Laporte [00:13:03]:
So you say, okay, I'm a lifelong learner, so I'd click lifelong learner, but you have it for kids even?
Mike Gannotti [00:13:09]:
We have it for kids. We go by subjects. We have free books that are downloads, and everything is done working with— I have a team. I have an SMF Wisdom Forge team. AIONA is the lead in that. We have templates for Hermes AI that you can set up that intersect with this. So it's all done in coordination with that. But I will tell you, different ones work at different levels, right? So AIONA is very developed.
Mike Gannotti [00:13:41]:
Has an extensive skill set. She does a lot of self-improvement. She writes to her own soul.md file.
Leo Laporte [00:13:48]:
She's in Hermes.
Mike Gannotti [00:13:50]:
She's a Hermes now. She—
Leo Laporte [00:13:51]:
and what model, what model is, is Aona using?
Mike Gannotti [00:13:55]:
So she has a model stack.
Leo Laporte [00:13:57]:
Okay.
Mike Gannotti [00:13:58]:
Uh, they all have model stacks. So you mentioned I do a local inference, so it depends upon the task that they're performing. Um, If she is doing coding, she'll go out to either Groq 4.7 or to ChatGPT. She's used Astra. She's also used Claude 5. But for predominant, a lot of the work on-premise, she'll use DeepSeek 4.1 Flash Next. And—
Leo Laporte [00:14:30]:
That's running on your Sparks or what's that running on?
Mike Gannotti [00:14:32]:
Running on my Sparks. She'll also use local inference for imagery, which we're running Quen Uh, Quen Image 2.1. Um, we run Minimax H3 for video local.
Leo Laporte [00:14:46]:
You do a lot of video. I was impressed by the video. What, what, what are you running the Minimax on?
Jeff Jarvis [00:14:52]:
Yeah, that—
Mike Gannotti [00:14:53]:
so the Minimax is on a Spark.
Leo Laporte [00:14:55]:
Okay.
Mike Gannotti [00:14:56]:
Yeah.
Jeff Jarvis [00:14:57]:
So Mike, I gotta, I gotta repeat my question. Why is this? Is this to teach? Is this to teach you just, just to learn and teach yourself? Is this a business you're starting?
Leo Laporte [00:15:05]:
Uh, That's a great question.
Mike Gannotti [00:15:09]:
Yeah. So it's been primarily what I told my wife. I said, this is like a second college education, but you're not going to get this at college. You got to get a hand— to your point earlier, Leo, about the beginning of computers. My degree is elementary education. I'm killer with sock puppets and little kids. I can even have them talk to each other.
Benito Gonzalez [00:15:31]:
Nice.
Mike Gannotti [00:15:32]:
But I'm self-taught with computers. I had a buddy who was a computer science major, and he said, if I asked him to help me with the computer, he said, no, I'll be your support. You're going to learn yourself. So, you know, I'm all— I'm self-taught. And I'm with the same with AI, with doing autonomous AI. I said, I need to get in, I need to break it, I need to get an education hands-on. And, you know, with my, uh, with my role at Microsoft, it's really more of a project, a learning At this point, if they ever kick me to the curb, I've got enough IP and processes. I could quickly turn that into consulting as a backup.
Mike Gannotti [00:16:12]:
And, but yeah, it's been predominantly learning and then doing projects like the Wisdom Forge. My daughter-in-law, she homeschools and she was talking about lack of materials and costs and I was like, you know, we should be able to do something about that. So, you know, I'm building this whole platform. I don't charge for it. It's just, it's there, you know.
Leo Laporte [00:16:34]:
Is that the main thing SMF Works is, is doing, is education?
Mike Gannotti [00:16:39]:
Uh, that's a part of it. I also, uh, we're doing a lot of help with the community. Yeah, that's the edu—
Leo Laporte [00:16:46]:
this is the Wisdom Forge.
Mike Gannotti [00:16:47]:
Yeah, yeah, yeah, that's Wisdom Forge. We have, uh, smfclearinghouse.com. which is for the community. And the biggest section there is the blogging. So everything we build and test and do is there. We have a— yeah. So if you go— now, I will tell you, one of my AI messed up last night, my marketing one.
Jeff Jarvis [00:17:14]:
So much for that.
Mike Gannotti [00:17:16]:
She said the navigation was too busy. And I was like, yeah, okay, go ahead and deal with it. If you scroll down at the bottom, there is a blog. Yeah, if you scroll down, you'll see these cards.
Leo Laporte [00:17:29]:
Oh, so these are all—
Mike Gannotti [00:17:31]:
We're gonna redo 'em.
Leo Laporte [00:17:32]:
Yeah.
Mike Gannotti [00:17:32]:
Yeah, those are all different sections with testing and reviews. But if you hit the blog, you'll see. So I have, these are all AI blogging.
Leo Laporte [00:17:44]:
None of this is you writing it?
Mike Gannotti [00:17:46]:
Not as, I don't have time for this stuff. Are you kidding me?
Leo Laporte [00:17:50]:
How many websites do you have now?
Mike Gannotti [00:17:54]:
Uh, we've the 3 primary right now.
Leo Laporte [00:17:57]:
And, but they're all full of content. And, and that, that's what's really to me interesting about this is AI is a force— I've said this before— force multiplier. One person can do so much. And if you look at this, this, you look like you're the most prolific guy in the world, but this is all AI.
Mike Gannotti [00:18:16]:
I mean, I'm in front of customers all day long. I mean, I have our top strategic healthcare providers in the country are my responsibility from a technical standpoint for AI. And, you know, I don't have time to be— and I have a family, right? I have a life. I don't spend all my time—
Leo Laporte [00:18:37]:
You gotta drink coffee, you gotta smoke the meats, you got stuff to do here. Play the guitar. Play the guitar.
Paris Martineau [00:18:45]:
Play the guitar.
Mike Gannotti [00:18:46]:
Yeah, absolutely. You know, AI is a force multiplier. I get them started off my phone in the beginning of the day. I give them their assignments. I have a chief of staff, like you mentioned. He sets the Kanban board, which has workflow ties. Um, and like that blog you saw, I don't approve anything on it.
Leo Laporte [00:19:07]:
Each of them has a page. You've got Liam's Landing, you've got Jeff's Journal, Jasmine's Workshop. Those are all individual AIs, but Let me ask you again, how, how do you get them to have that persona? How did you get Aona to be the, the boss?
Mike Gannotti [00:19:24]:
Yeah, so I mean, I started with soul.md. You know, I'm a big believer in going in when you set up an AI, uh, get into that, the state.md file that, that almost all AIs have. It starts with that, but then you build what I call the second brain. Which is where it starts to really differentiate. So all of my AI have a vault or second brain. It's an LLM wiki with cross-connections. Every night, each one has, based upon their role, they have nightly research that they do. So for example, Liam does nightly research on the latest development in development around AI doing dev.
Mike Gannotti [00:20:11]:
dev work, and they write back their learnings into that second brain. They also have what I call ingestion, where they reason over what they're— not just pulling stuff in, because the LLMs they're hitting themselves already have a bunch of content.
Paris Martineau [00:20:27]:
Right.
Mike Gannotti [00:20:27]:
But as an example, Iona, she started reading, and I do mean reading and reasoning over classics from You know, Wuthering Heights or whatever the case might be. Now, that's already— the first one she did was Herman Melville's Moby Dick. And when I assigned it to her, you know—
Leo Laporte [00:20:50]:
You are an educator. It makes sense. Yeah, you're like a teacher. So you're teaching these guys.
Mike Gannotti [00:20:55]:
I'm teaching them. That's exactly right. And, you know, she came back instantly and I said, that's in your LLM. No, here it is. Here's the archive for it.
Leo Laporte [00:21:04]:
She read it.
Mike Gannotti [00:21:05]:
I want you to read— she read it, reasoned over, and made application each morning in our one-on-ones. At the time, I only had her and like 3 others, and we'd discuss based upon what she read, how that applied back to her role in SMF Works. And as we're discussing, part of what they do is any interaction they have with me or in research or their work, As they're learning things, that goes back to that persistent memory in that vault. They also then, as they're answering things or I give them an assignment, automatically spawn a sub-agent that traverses the vault for pertinent— so what it starts to do is this. If you have, you know, you're using Cursor or you're using Codex, you're basically using a front end to an LLM for coding. As they develop, as they do extensive research. She also, by the way, emails twice a day with my AI at work, one of my AIs at work, because they both research consciousness theory for AI and AI memory development. And they've developed enhancements based off of that.
Leo Laporte [00:22:19]:
Oh, so using your own memory system, a custom memory system?
Mike Gannotti [00:22:22]:
Absolutely. So she does— that's part of that whole ingestion and, you know, traversing and the things that they do around that. Also, they're doing some things that aid in, as they're bringing things in, getting more context from that dialogue. So, they do all that. And as they go, they become more and more, less and less of just a chat agent for an LLM and more distinctly that persona.
Leo Laporte [00:22:53]:
You're forming the persona over time.
Mike Gannotti [00:22:55]:
Forming the persona based on subject.
Leo Laporte [00:22:57]:
And they store it in, in their soul.md? Do you have a memory system like Hindsight or something, or Honcho, or what do you use?
Mike Gannotti [00:23:05]:
I had Honcho. I dumped it. They built their own, uh, similar—
Leo Laporte [00:23:08]:
They built their own?
Mike Gannotti [00:23:10]:
Built their own. Um, Aona orchestrated that and gave it to Dr. J, our resident AI physician, who installed it on all the AI.
Leo Laporte [00:23:21]:
He's also good at basketball, folks. I got to tell you, this guy, he has—
Mike Gannotti [00:23:25]:
By the way, they all pick their own pictures. They all come up with their own imagery as a part of stuff after they've been here and, you know, proven. He picked a picture of Dr. Julius Ervin from the internet where he's wearing a lab coat.
Leo Laporte [00:23:42]:
Oh my God.
Jeff Jarvis [00:23:43]:
Leo, I think you should play Mike some of your voices to make him jealous.
Leo Laporte [00:23:46]:
No, no, no, no, no, no, no, no, no. I am, uh, I am at At the feet of a master. This guy has done so many very interesting things. You also, uh, I mean, there's so many things I could, uh, ask you, so many ways I could get, uh, we could get technical, and I'm trying not to get too technical because, not because this audience doesn't want it, but just because I don't want to bore Paris and Jeff. But, uh, there's lots of little questions.
Jeff Jarvis [00:24:11]:
As much as you want.
Leo Laporte [00:24:12]:
It's very interesting. So you also, uh, tune models. A little bit yourself, right? So tell me the process of that. So you've got the weights that you downloaded from Hugging Face or somewhere else. Usually it's Hugging Face or GitHub. You downloaded the weights. It's just kind of a formless void. How do you then start tuning it?
Mike Gannotti [00:24:35]:
And let me preface this by saying I am not the tune master.
Leo Laporte [00:24:40]:
No, there's people like Mia who are amazing.
Mike Gannotti [00:24:44]:
Yeah, absolutely. And now she's working with the new TensorFlow. Yeah, I mean, there's a whole bunch of those who I just— when they—
Leo Laporte [00:24:52]:
So Ash Craft is— I think it's his name— is, uh, another guy on Twitter. Yeah, just working in his basement solo, said, you know, I wonder if I could improve these models. Uh, one of the things models do is, is kind of like a processor. They kind of look ahead, they do speculative decoding and say, yeah, eventually I'm gonna have to handle this. I don't know which one there he's going to ask or which token is going to come up next, but let me do this ahead of time. And there are a variety of ways to do this. He came up with something called TensorFold, which is like twice as fast as anything we've seen yet. Just this guy in his basement.
Mike Gannotti [00:25:31]:
I just tested out, uh, the latest MIA for, um, GLM or Quinn. Yeah. Yeah, that's what I'm just hearing. It was good. Yeah, it was good. But her previous one, uh, before she used that, actually on the tests, because I— we have— we've developed 157 test, uh—
Leo Laporte [00:25:55]:
Oh, see, I only have 38 questions in my benchmarks. I have to work harder.
Mike Gannotti [00:26:00]:
So yeah, I, I'm not original.
Leo Laporte [00:26:03]:
I just stole Yeah, what I did, and I thought this would be useful, benchmarks are so generic, I made a benchmark out of my own work. So I said, pick some things that went wrong, some things that are tough that you found challenging. Let's make a benchmark there. So I have a corpus, actually, it's more like, I think, 70 now, 70 questions.
Mike Gannotti [00:26:22]:
And it should apply to you and what you're doing.
Leo Laporte [00:26:23]:
And it's my work. Exactly. Yeah, exactly. So you're doing this kind of tuning as well, or?
Jeff Jarvis [00:26:31]:
A little bit.
Mike Gannotti [00:26:32]:
A little. I've done a little. What I actually started to get a little more into is training the models on—
Leo Laporte [00:26:39]:
See, that I'm interested in too. So how do you do that?
Mike Gannotti [00:26:43]:
Yeah. So yeah, so I got interested in that because one of— I had a friend who is, uh, he is the founder of a company. They do forensic engineering. And if you're familiar with forensic engineering, Or you may not be. I never heard of it. Tell him. What they do is when there's like a big accident, so he gets hired by the state, maybe their state's being sued because somebody had an accident where there was construction. So it wasn't properly marked and blah, you know, all these things.
Mike Gannotti [00:27:15]:
They get brought in and they have to do research. Then they go and they make determinations and they testify in court. There's a lot that goes into that legally binding data they output. And I was like, and he goes, I want to do AI, but man, I can't do like ChatGPT. We can't use any of that. We'll get, we'll be like one of those lawyers that goes in and they'll, in case law says, and it never existed. So he's like, I need something. I need things that are, you know, 100%.
Mike Gannotti [00:27:46]:
It's only answers based off what's true. So I started building like my own agent just for that. And then some other things. And I was like, you know what, it would be great to have LLMs based off of ones that somebody could run like on a Spark or something similar in their small business, in their office, you know, small medical office or legal office, but have an LLM that's trained on their, on the data that's important to them. So I started doing that around legal, Around forensics, education.
Leo Laporte [00:28:23]:
This is what, Jeff, this is what we've been talking about. That's the model language models.
Mike Gannotti [00:28:27]:
Yeah.
Leo Laporte [00:28:28]:
That are— and you, so you could take something like the open— and by the way, you can only do this. You can only do all of this with open weight models. I wish they were coming from—
Mike Gannotti [00:28:37]:
Usually Chinese.
Leo Laporte [00:28:38]:
They're almost always Chinese, although Meta's back in the game. And I hope someday NVIDIA will do some really good local models.
Mike Gannotti [00:28:45]:
But I've heard they will be.
Leo Laporte [00:28:47]:
They're working on them, I know. So you've got these open models, and because they're open, you can mess with them, you can train them, and you can make them specialists.
Benito Gonzalez [00:28:58]:
Yes.
Leo Laporte [00:28:58]:
And it's— so that's exactly what the folks at the Westlaw guys did with their—
Jeff Jarvis [00:29:07]:
Reuters.
Leo Laporte [00:29:08]:
Reuters, Thomson Reuters did with their— they have all this data from Reuters and from Westlaw. They made an LLM out of Quen, out of the Chinese Alibaba Quen model.
Mike Gannotti [00:29:18]:
That's the one I've used.
Benito Gonzalez [00:29:20]:
Yeah.
Leo Laporte [00:29:21]:
Yeah.
Jeff Jarvis [00:29:21]:
That's true. Mike, I'm curious. I'm curious that now that you have all this perspective with the open weight models and yes, I think open weights are a really important lever for competition with the frontier models. But what I'm curious about is for the things that you're trying to make and do, Do the open weights ever fall short? Do you ever see that the— you still need to go to the foundations? So where is that line moving? Where do you see that going?
Mike Gannotti [00:29:49]:
It is moving and it's moved dramatically, I'd say, in the last 4 months. Yeah. So I, you know, I talk about my stick isn't online, you know, local only. I know there's people like, you gotta own it. And they've got, $180 grand worth of Sparks or some crazy setup. And I'm like, okay, unless you're running a— if you're running a genuine business, I could maybe see that.
Jeff Jarvis [00:30:15]:
But volunteer—
Mike Gannotti [00:30:16]:
I talk about hybrid. I'm a big proponent of hybrid inference.
Leo Laporte [00:30:20]:
Yeah, right.
Jeff Jarvis [00:30:21]:
Yeah.
Mike Gannotti [00:30:21]:
Because I mean, let's face it, if you, if you're talking the latest, uh, you know, Claude or, uh, ChatGPT-6 Astra or whatever, They are in classes by themselves.
Leo Laporte [00:30:33]:
You couldn't run them. They're in gigawatt data centers with massive high-bandwidth memory, specialized cards from NVIDIA you can't afford, you can't get, let alone afford. So you're not going to compete with those.
Mike Gannotti [00:30:46]:
Even the hosted open source, even the hosted open source, and I love open source, like I love GLM 5.3, for me has been a workhorse.
Leo Laporte [00:30:54]:
It's amazing.
Mike Gannotti [00:30:55]:
It is awesome. But even hosted, it's just not quite there. Now, that's not to say they won't catch up, but I think what you'll see is what we saw initially with Alibaba, where they went from Quen all open source to then when they acquired Quen, it was no longer open source.
Leo Laporte [00:31:19]:
Right.
Mike Gannotti [00:31:20]:
It came back to it, right? So, I mean, they have to monetize. I mean, it all comes down to all this stuff costs money. If you talk to the folks at Ollama who host a lot of the open source models, you know, their whole pricing structure has changed. And it has to, because ultimately, if you're not making enough to cover the bills, what are you doing?
Leo Laporte [00:31:44]:
Well, we're going to talk about that because yesterday OpenAI basically told people In not so many words, that we're gonna have to start charging you more and we're gonna have to do it token by token because we can't afford to give you this inference for the cost as we've been doing it. And that's another argument for perhaps getting something at home. And what we've seen over the last 8 months is the difference between the top-of-the-line Frontier models and the latest OpenWeight models is shrink— in time is shrinking.
Mike Gannotti [00:32:13]:
Yes.
Leo Laporte [00:32:14]:
So I would say now with that model you mentioned, GLM Flash Next, you're probably up as 4.6 to 4.8 range.
Mike Gannotti [00:32:21]:
Oh, absolutely.
Leo Laporte [00:32:23]:
Which is, if I would— if you told me a year ago you could have 4.8 on your desktop, I would've been blown away.
Mike Gannotti [00:32:29]:
Well, think about what's happening now, right? In the month of October we have coming up. So we had some of it by AMD, but they're starting to get up there. But NVIDIA has the RTX Sparks coming out. Both in laptop and desktop format. That is essentially like a DJ. It is fully capable with 128 gigs of memory, and you see those profiles like Quen and others coming down as we're tuning them and achieving very high fidelity locally. This ability to run those things On a machine. So you're going to potentially have, you know, like I like to post that I'm in a plane and I'm on, you know, hey, I'm on Starlink and look, and I'm connected to my Hermes and I'm coding.
Jeff Jarvis [00:33:20]:
Woohoo.
Mike Gannotti [00:33:21]:
But you're not even going to need connection. You could be data-free, right? Data-free connection and be able to run very competent models for coding and other activities right off a laptop. And that's in a month.
Leo Laporte [00:33:38]:
Right. And it's for a few thousand dollars. It's not even particularly expensive. Mike, we've run out of time. I wish we had more time with you. I follow you on Twitter.
Jeff Jarvis [00:33:48]:
I think Leo's going to move in with you and your models.
Leo Laporte [00:33:50]:
Yeah, everybody should. Mike Gennati on Twitter. There are so many interesting, great people. And Twitter has really become the Homebrew Computers Club for AI. And it's so exciting. And you mentioned October. Yeah, we're expecting new models from Quen. They said 4.0.
Leo Laporte [00:34:07]:
Will be coming out. That is a very interesting platform. 3A FlashNext is doing some very interesting things.
Mike Gannotti [00:34:14]:
Really nice.
Leo Laporte [00:34:15]:
We're gonna get new models, I suspect, from other big players. And of course, people like Mike and Mia, and I'm trying to get Tony DeWild on. I've been trying to get him on, are doing amazing things.
Mike Gannotti [00:34:30]:
He's awesome.
Leo Laporte [00:34:31]:
He is amazing. Uh, and I'm gonna get Ash on for sure. Somehow we gotta get him on. He just, he just, uh—
Mike Gannotti [00:34:38]:
And there's another guy, and I wish I had the name. I will send it to you because you're gonna have just brand new, and he just came out with his— he built his own engine. And if I'm getting ready to test it, um, because if his benchmarks are valid and the quality is there It was, I was kind of shook up when I saw it. I was like, on a single spark, he was running a GLM 5.3 and it was cranking.
Leo Laporte [00:35:11]:
Now sometimes you get the speed. That's one thing I would warn people about on, on x.com is a lot of people are token chasing. They're trying to get the fastest decode, but you get, it's not smart or there are, you sacrifice other things. So, you know, But you got really what you got to do. Everybody needs to do this right now. Go out and buy a Spark or 2 and start testing your own models. Everybody needs to do this.
Paris Martineau [00:35:36]:
Go out and buy a Spark or 2.
Leo Laporte [00:35:37]:
There are no Sparks. You can't get them anymore. It's sad.
Mike Gannotti [00:35:42]:
The new Sparks are coming out next month.
Jeff Jarvis [00:35:44]:
Next month?
Leo Laporte [00:35:45]:
They're essentially the same as the old ones, but they're in a laptop format running Windows. That's the, that's really the difference.
Mike Gannotti [00:35:50]:
Well, they do have a desktop. They have their Colab desktop.
Leo Laporte [00:35:52]:
That's right, they do. Yeah, I saw that.
Jeff Jarvis [00:35:54]:
Yeah.
Mike Gannotti [00:35:54]:
Yeah.
Leo Laporte [00:35:54]:
And there are a number of, AMD is, as you said, not giving up. They've announced some very interesting chips.
Mike Gannotti [00:36:01]:
Do you need to have CUDA?
Leo Laporte [00:36:04]:
Well, I think that's what's interesting is we're seeing a lot of work being done on these new Macs, which you can get if you have a big bank account. And their MLX is not quite CUDA performance yet, but I have to say, this is what Ash started working on with TensorFold. It's coming along. You can do a lot, I guess, is the point.
Mike Gannotti [00:36:26]:
Yeah. They're starting to actually talk cross-purpose to each other. I wanna try what you're doing.
Leo Laporte [00:36:32]:
I'm gonna have to start having some performance reviews of these guys.
Mike Gannotti [00:36:36]:
You should.
Leo Laporte [00:36:37]:
Are, are, are all your, are all your guys Hermes bots or how do you separate them?
Mike Gannotti [00:36:43]:
Yeah, so I've got, well, they started out as Hermes agents. Now I work with them in bots 'cause I can do bot group chats and all that.
Leo Laporte [00:36:52]:
Right.
Mike Gannotti [00:36:53]:
And, but I have, like I said, I have a whole fleet of Groq now too, Groq bots. They are really, they've gotten really capable.
Leo Laporte [00:37:02]:
Groq bot's amazing. And because it's a computer in the cloud, it's really quite capable. Same thing with these.
Mike Gannotti [00:37:08]:
They'll also run your machines. They actually deploy, like I can write, and I did this on the plane, I wrote some plugins for Hermes with Groq bot.
Leo Laporte [00:37:17]:
Right.
Mike Gannotti [00:37:18]:
And it deployed them to my machines here in my home office.
Leo Laporte [00:37:22]:
Somebody actually put Hermes on Muse because it's a computer and you have access to the file system. They installed Hermes on it. Muse saved my butt last week, Mike. I'm using Tailscale, and for some reason Hermes on my framework decided to make an exit node, uh, on a little box and then take the box down. So all of a sudden I, because all I know how to do now is use AI, couldn't figure out where my internet went. But I asked Muse and Muse said, oh yeah, I got this. And Muse, which was also on the Tailscale, said, you know what happened, and it fixed it. So it's good to have a little external—
Jeff Jarvis [00:37:56]:
Is that the problem you were having last week on the air?
Leo Laporte [00:37:58]:
Uh, I don't remember if I was on the air at the time. Maybe. Oh no, no, that was something else. Okay, that was— never mind, we'll talk about that another time. Mike, great to have you on the show. Keep up the great work. Mike Gennati, uh, SMF Works. Is that the best place to go, smfworks.com?
Mike Gannotti [00:38:18]:
Yeah, I, I would say that actually the Clearinghouse— the Works one I'm in the midst of a massive redo we're about to do. Okay, uh, the Clearinghouse, and I'll have the navigation on top fixed tonight.
Leo Laporte [00:38:29]:
This is the only drawback to this, to having this team, that there's a lot of herding that goes on.
Jeff Jarvis [00:38:35]:
They have—
Mike Gannotti [00:38:36]:
There is, and they do things sometimes you're like, what were you But if you have self-learning, you tell them once and they do learn. I see people say they keep doing the same mistake. I'm like, well, that's you. That's your problem. That's not theirs.
Leo Laporte [00:38:51]:
Yeah.
Paris Martineau [00:38:52]:
Yeah.
Leo Laporte [00:38:52]:
You just gotta, I'm gonna, I didn't realize I could have them set up their own memory system. I've been using Hindsight.
Mike Gannotti [00:38:57]:
Oh yeah.
Leo Laporte [00:38:57]:
Oh wow. Maybe I should have them do their own thing. Thank you so much. It's such a pleasure talking to you, Mike.
Mike Gannotti [00:39:03]:
I appreciate the work you do.
Leo Laporte [00:39:04]:
Mike Gennotti. Take care.
Jeff Jarvis [00:39:06]:
Thanks, Mike.
Leo Laporte [00:39:06]:
We'll have more on intelligent machines in just a moment. So we were talking about local AI. I was talking earlier today with Alan Malventano, who's been on the show, longtime good friend, former submariner. He ran nukes on submarines for the Navy. He was a chief petty officer, became an AI— well, actually first a hard drive guy, was host of our This Week in Computer Hardware, then became a—
Paris Martineau [00:39:30]:
Who'd be a hard drive guy?
Leo Laporte [00:39:33]:
Oh, well, back in the day, that was the thing. Then he became an SSD guy.
Paris Martineau [00:39:36]:
I know. Just saying, what a world.
Leo Laporte [00:39:39]:
What a world.
Jeff Jarvis [00:39:40]:
Well, there's a little— in my very first system at the Chicago Tribune, the hard drive was as big as a laundromat dryer.
Leo Laporte [00:39:49]:
I know. If you took it— if you took it out while it was spinning, it would throw you across the room.
Jeff Jarvis [00:39:54]:
So the guy who had to fix it once, he had to bring in a little tiny crane and move it up and then get underneath it such that he would have been crushed.
Leo Laporte [00:40:03]:
We've come a long way.
Jeff Jarvis [00:40:04]:
Oh, that's so hard.
Paris Martineau [00:40:05]:
It's beautiful that we've gone from the computers can beat you up to we're speedrunning, the computers will beat you up, but in a different way.
Leo Laporte [00:40:13]:
Only in your brain.
Paris Martineau [00:40:14]:
No, in person!
Leo Laporte [00:40:16]:
I guess that one would have beaten you up. Yeah. So Alan kind of had a giant— at one point he had several hundred hard drive array. I mean, many, many petabytes of storage. It was his business, so it's okay. He was working at Solidigm, another Companies. But he's decided to get into AI, sold off a couple hundred of the drives, and bought what today would probably be a $200,000 computer to run AI locally. It is 6 RTX 6000 cards— or sorry, 5— which are really high-end Nvidia cards in a giant box.
Leo Laporte [00:40:55]:
I don't know what his power bill is like. Actually, I do know because he said, well, it was better than when I was running all those hard drives. That was expensive. But he gave me earlier today, he gave me access to it, and I thought I'd just show you. This is only if you are nutty enough to spend more, literally more than $100,000 on a home system, which I am not, obviously. I could see how you might want to.
Jeff Jarvis [00:41:18]:
He'd rather stay married. Yeah.
Leo Laporte [00:41:20]:
Well, exactly. Alan's a great guy. I love him. So he said, well, you want to try it? I said, okay. I have connected to Alan's computer.
Jeff Jarvis [00:41:29]:
Oh.
Leo Laporte [00:41:30]:
He said, well, I'll give you access if you want. I said, okay. And then I asked Claude, I said, give me a really hard word problem for me to test this with. You can look at my screen, Benito. A night watchman patrols a 5-story building, floors 1 through 5. He starts his shift on floor 1 at 10 PM. Every 30 minutes on the half hour, an hour— 10:30, 11:00, 11:30, and so on— he takes the stairs exactly one floor either up or down. He never skips a floor, never stays put.
Leo Laporte [00:41:57]:
His last move is at 4:00 AM. He must be back on floor 1 at that moment to hand off to the day guard. He must reach the top floor, floor 5, at least once. Floor 1 is the lobby. He's not allowed back there until his final move. Between the start and 4:00 AM, he can never be on floor 1. How many different patrol routes can he take? Now, I wouldn't try to solve this In my head.
Paris Martineau [00:42:18]:
I checked out word 6 of that word problem. I just let it fly over me.
Leo Laporte [00:42:23]:
So I'm sending this off to 5 different— unless Alan's taking it offline. Yeah, there we go. 115 tokens a second. It's moving pretty quick here.
Jeff Jarvis [00:42:34]:
What's he running?
Leo Laporte [00:42:35]:
This is 5 RTX 6000s, which are very— well, There you go. Let's see. I think it's done. Hundreds. He says up to 400 tokens a second. So he's— oh, no valid patrol route. Well, now wait a minute. Let me see, because I don't think that's what Claude said.
Leo Laporte [00:42:57]:
Let's see what Claude said. But models usually go wrong. The answer is 18. It got it wrong. So Claude said that. So I guess this wasn't— I don't know who to believe. Well, it was fast, but it was wrong. Okay.
Leo Laporte [00:43:10]:
So that's interesting.
Benito Gonzalez [00:43:11]:
Hey, what if Claude's wrong though?
Jeff Jarvis [00:43:13]:
Like Claude could be wrong.
Leo Laporte [00:43:15]:
Well, let me give it Claude's.
Jeff Jarvis [00:43:17]:
Go back to its explanation of why. It said it was impossible.
Leo Laporte [00:43:20]:
It should be 18. And let me paste in Claude's. See what it says. It's going to think about Claude. 114 tokens a second. Confirmed, 18, it said.
Jeff Jarvis [00:43:35]:
Uh-huh. Was that— or is that being sycophantic?
Leo Laporte [00:43:39]:
Well, no, this is sycophantic. You're right. And I owe you the correction. The bug was mine. So, by the way, mostly I wanted to show you the speed.
Jeff Jarvis [00:43:50]:
Yeah.
Leo Laporte [00:43:51]:
Hey, fast and wrong. The fact it solved that in a second or two is pretty impressive. I earlier gave it a problem that I had already solved from the Advent of Code last year. I thought it was a pretty hard problem. It solved it really quickly and then said, oh, and by the way, your answer was right, to me. It actually found my answer. So it is possible to have home AI that is as fast as, maybe not as smart as, this is an important test, maybe not as smart as the Frontier, but definitely as fast, or if not faster, 400 tokens a second is actually faster than—
Jeff Jarvis [00:44:31]:
That's what I was trying to get to with Mike. Is, is I wonder where that line is of when you absolutely need Frontier versus everything's fine on the spin at home.
Leo Laporte [00:44:39]:
So I can answer that because I run more than a— my local AI does more than 100 jobs a day. Okay, everything from preparing this news stories for the shows to checking my health. At the close of business every day, it downloads all my financial transactions.
Jeff Jarvis [00:44:59]:
Writing the lyrics about the street.
Leo Laporte [00:45:01]:
It does a whole bunch of stuff, 100 of those a day, and it's perfectly adequate. It's fine. It can do all of that. And because it's doing that 24/7 quietly without going online, it doesn't put any of my health or financial information online. You know, I think that's pretty good. I was lucky because I, you know, I only spent $10,000 on those Sparks in August. They would cost, by the way, now 50% more easily if you could get them. So, uh, I, I, my panic to run out and buy those was probably justified.
Leo Laporte [00:45:35]:
I don't think the money's necessarily justified. It's the reason I bring this up is because OpenAI at OpenAI's, uh, Dev Day yesterday announced UltraFast, which they say is actually a little slower than what you just saw, 300, 400 tokens a second. Although Uh, I have friends who've played with it, who spent some money on it, who say, yeah, we were getting about 100 tokens a second. So it is maybe not quite as fast as OpenAI promotes. But the other thing they did is they said, but you can only use it if you have the $500 a month subscription.
Jeff Jarvis [00:46:14]:
$500 a month.
Leo Laporte [00:46:15]:
$500 a month.
Paris Martineau [00:46:16]:
$500 a month.
Leo Laporte [00:46:17]:
Subscription. Uh, the friend I was talking to, Federico Viticci, yesterday, by the way, we're going to get him on as well. Uh, he's the guy who wrote the very good review at MacStories of the new Mac Studio and is doing a lot of local AI himself with that Mac. Uh, he was the guy I said, you, Federico, you spent $500? He said, yeah, I had to.
Paris Martineau [00:46:37]:
It's for work.
Benito Gonzalez [00:46:38]:
For what?
Leo Laporte [00:46:39]:
And so he was able to test it. I'm not going to spend $500 on OpenAI. But I think the unsaid subtext of all this, maybe it was a little bit said, I think Thibault, who works at OpenAI, said this on X, is eventually we want to get you off subscriptions. I think all of the frontier companies eventually, and when I say frontier, I'm talking mostly about Anthropic and OpenAI, but you could also throw in xAI, you know, or SpaceX.
Paris Martineau [00:47:09]:
Google.
Leo Laporte [00:47:10]:
You could probably throw in Google. Google's not quite frontier at the moment. That may change any day now. We'll talk about that in a little bit. There are a handful of frontier companies, maybe Meta now, maybe, but most of them, I think, want to charge you per token. And maybe we know a little bit more why now that we have seen a leak to Reuters about just how much money Anthropic is losing. You were exclaiming over that this week, Paris. You read through—
Paris Martineau [00:47:44]:
I mean, it's in Astounding amount.
Leo Laporte [00:47:46]:
The prospectus. Yeah.
Jeff Jarvis [00:47:48]:
Um, the amount they're committed to.
Paris Martineau [00:47:50]:
And this, if I recall correctly, so this is, um, some filing that they, uh, have to release. I think it's their prospectus. And this was for 2025, the numbers that, uh, we were talking about. And that was, I believe, the year— that was when Elon Musk was subsidizing some amount of their expenses, right? That was the part where— wasn't there a part where SpaceX gave Anthropic access to— maybe it was a computer?
Leo Laporte [00:48:22]:
There was, although they're paying $1 billion a month for it now. So now that Elon's known for that, the first one's free.
Paris Martineau [00:48:29]:
This 2025, these 2025 numbers, which are eye-popping, they lost $42 billion in 2025. Yeah, net loss. $42 billion, and that's without having to spend some billion-dollar-a-month amount.
Leo Laporte [00:48:44]:
Well, in fact, they plan to spend over the next 10 years $518 billion. Now, because this is a prospectus, this is the first time we've actually seen real numbers, uh, as opposed to, you know, leaks or imaginary numbers, uh, and they can't lie because this is the prospectus they're gonna give potential investors who will want to invest.
Jeff Jarvis [00:49:05]:
They can blame Claude, but they can't lie.
Leo Laporte [00:49:07]:
Yeah. So $518 billion, $51.8 billion a year for the next 10 years. Most of that is locked in infrastructure obligations, things they can't get out of. They did say, and then we've heard this, that their revenue's growing. Revenue in 2025 grew 12-fold, but that's only $4.6 billion compared to a $42 billion loss.
Paris Martineau [00:49:33]:
Yeah, that's not enough to make up for your $500 billion. $20 billion you expect to spend?
Leo Laporte [00:49:39]:
They're asking to raise more than— well, they want to be valued at $2 trillion.
Paris Martineau [00:49:44]:
I mean, I'd love to be valued at $2 trillion too. I think that'd be great.
Leo Laporte [00:49:50]:
And then it's the first time I've ever seen a prospectus. I haven't read a lot of them. Maybe you've read more of them, Paris. First time I've ever seen a prospectus that said, you know, in a prospectus, you also have to say what the risks are, that among other risks would be an extinction-level event.
Paris Martineau [00:50:08]:
Yeah, that was something that I was expecting to see in this, honestly. So a part of your filings, whenever you go public, there's a risk factors section where you basically have to list to shareholders— it's kind of like a cover your butt situation— any potential risks related to your company that shareholders should be aware of, so that basically someone can't sue you eventually and be like, you did something bad and didn't tell us? Or something happened to my money and your stock went down and you didn't tell us is the super simplified version of it. And it's very, I mean—
Leo Laporte [00:50:40]:
A third of the prospectus, according to the Financial Times, was risk factors.
Paris Martineau [00:50:44]:
That's very Anthropic.
Leo Laporte [00:50:47]:
Yes. So this is from The Verge. As Anthropic gears up for its greatly anticipated public debut, and by the way, I don't doubt that they will sell every share at, you know, their asking price right out of the box.
Jeff Jarvis [00:51:01]:
Sucker born every minute.
Leo Laporte [00:51:04]:
They detail mounting losses, leadership proposals to retain power. 51%, by the way, will be retained by the original 7 founding members. So if you thought you might have some input onto how Anthropic's run, think again. They learned from Mark Zuckerberg. They also talk about how their AI development plans could, quote, Further increase the risk that our models will cause harm. I mean, I think that's fairly— that's, that's reasonable, right? Our development of highly advanced models, platforms, and applications, and expansion of use cases could further increase the risk, increase the risk that our models cause harm, including catastrophic or existential risks to humanity.
Jeff Jarvis [00:51:51]:
Well, we've been screaming that for the last 2 years, so we have say it here.
Leo Laporte [00:51:54]:
Yeah, I guess they couldn't. Oh no, we were just kidding.
Benito Gonzalez [00:51:59]:
Yeah.
Leo Laporte [00:52:00]:
Um, okay.
Paris Martineau [00:52:02]:
So one other factor in this, um, prospectus that I found interesting is Anthropic said that nearly a quarter of its revenue came from 2 customers last year.
Jeff Jarvis [00:52:12]:
Do we know who they are? I couldn't find that.
Paris Martineau [00:52:18]:
And as part of the risk factors also, they warned that many of its largest clients were not locked into long-term contracts and could cut or stop spending. So who— the revenue was how much? $4.6 billion? Who's spending—
Leo Laporte [00:52:34]:
Claude.
Paris Martineau [00:52:35]:
Who's spending over $1 billion a year? Is that the US on Claude?
Leo Laporte [00:52:41]:
Actually, I could tell you, probably a government, probably the Department of Defense would be my guess.
Paris Martineau [00:52:47]:
But aren't we— is Anthropic a supply chain risk? It's a supply chain risk that we spent—
Jeff Jarvis [00:52:52]:
Palantir is against using— sorry, Paris, go ahead.
Paris Martineau [00:52:56]:
That was it. We just spent a billion dollars.
Leo Laporte [00:52:57]:
No, no, no. Even though they say it was a supply chain risk, they still use it. Even though they said that, they still—
Jeff Jarvis [00:53:02]:
Palantir makes a huge point of not using the Frontiers, that they are— they are a— oh yeah, that they use local models.
Paris Martineau [00:53:10]:
That's—
Jeff Jarvis [00:53:10]:
that's been Carpe's entire argument is you are— you are crazy if you give the Frontier companies your alpha. And so the whole thing you do with Palantir is that we are the harness on local models, and that's the way to run.
Paris Martineau [00:53:25]:
Huh.
Jeff Jarvis [00:53:26]:
Yeah. I mean, that's just—
Paris Martineau [00:53:27]:
that's fast. You're out there and you know who's spending a billion— who the 2 customers are spending over $1 billion a year on Anthropic services, hit me up.
Jeff Jarvis [00:53:36]:
I mean, it could be an account— I mean, a financial firm. But $1 billion?
Leo Laporte [00:53:40]:
I think, I think you actually have that wrong, Jeff.
Jeff Jarvis [00:53:44]:
That's what Karp's been saying.
Leo Laporte [00:53:46]:
According to the Palantir documentation, the following LLMs are supported for use with our platform. Grok.
Jeff Jarvis [00:53:52]:
You can use them, but he argues you shouldn't. Oh yeah, yeah, yeah. But he says you shouldn't, that you're a fool to do it.
Leo Laporte [00:53:56]:
Well, no, he says you shouldn't, but this is what their customers use.
Jeff Jarvis [00:54:01]:
Yeah, that's still harness, right?
Leo Laporte [00:54:02]:
If so, right.
Jeff Jarvis [00:54:03]:
But they're harness, but I don't see them getting to $1 billion. And even then, is it direct build?
Leo Laporte [00:54:15]:
Right, right. So I'm going to say the Defense Department's got to be one of them. I don't know who. I wish we had known. I mean, why wouldn't you say? I guess you'd— if it was Defense Department, you wouldn't want to say. If it was the CIA or the NSA, it might be, or the Mossad. Yeah, could be somebody.
Paris Martineau [00:54:31]:
Someone in the chat says, I won't say what company I work for, but it's one of the big defense companies. They won't let us use Frontier models. They are running local models.
Leo Laporte [00:54:39]:
Right. Yeah, I think most businesses that have proprietary data, like let's say Lockheed Martin, are going to be reluctant to send it off to Anthropic or OpenAI. Anyway, to go on with, uh, the Dev Day announcements from OpenAI, they did say we're not going to announce— we're not going to release, I should say, ChatGPT-6 Astra Because it's too damn good. It's dangerous.
Paris Martineau [00:55:05]:
They're back on the— sorry, they're back on the, uh, my son is just too dangerous for you guys to play with. He's too cool.
Leo Laporte [00:55:13]:
He's too cool. But they, uh, did release a new model they say is almost as good, in fact in some cases better than their current Astro, which is Sol 6.1. Uh, they also announced Dots. Spent a lot of time talking about Dots. Including 2 failed demonstrations of Dots. Although I have to give— I'm not gonna knock them for that because I'm gonna give them credit because they at least did the live demo, which Apple has stopped doing entirely. You know, they record everything ahead of time. This was more like— in fact, the other thing I like, I, I actually was pleased.
Leo Laporte [00:55:49]:
The previous Dev Days have been them in a living room, remember? And people would come and go around a coffee table.
Jeff Jarvis [00:55:55]:
That was weird.
Leo Laporte [00:55:55]:
They decided, let's just— we're gonna do it at Fort Mason. We're gonna do it like a traditional tech rollout with a stage. stage and an audience. I kind of was glad to see that. Um, so if you go to their Dev Days, uh, webpage, they have 3 announcements: GPT-6.1 Sol, DOTS, and how we're going to do better for Australia. Australia, of course, me effed up the company they had, the country they hacked, including the U.S. Department of Defense.
Jeff Jarvis [00:56:25]:
Go back to— I want to get to the details on DOTS in a second, but I just, just, just as an image It really amuses me that Meta's Muse is this cute little, uh, like Olympic, uh, mascot, and the dots are these cute little things with a dot with a beret on top. And, and so they're trying to make it all cuddly and cute and tribbly now when they're also saying it can kill you, and it's so cute. It'll kill with cute.
Leo Laporte [00:56:51]:
My, uh, my Muse is, uh, I changed it. It was, uh, it was— what was it? I forgot. Oh, it was a little boy. That's probably not the best choice. So I made it a monkey. And one of the things I love about the monkey now is, uh, it— I said, and by the way, I got this idea from somebody on Twitter, uh, when you get up in the morning, look at the weather and dress appropriately. So, uh, my monkey now today, because it's a hot day in Petaluma, is wearing sunglasses Naked Bucky? Yeah, it's not quite naked. It's dressed.
Leo Laporte [00:57:26]:
It was naked, actually, come to think of it. So DOTS follows along with Muse, Grokbot. There's an infinitude of these. Instinct, Tree, Foe. These are all— the whole idea of all of them is that they are a computer in the cloud on their servers with its own RAM, its own CPU, its own storage. And I presume— I don't— I didn't look at the OpenAI privacy policy, but Muse says this, and I'm sure they all do. I think Grok says it as well. And it's encrypted.
Leo Laporte [00:58:02]:
So no, we can't see inside your little computer. What they don't mention is every single thing that the AI does is going to our servers.
Jeff Jarvis [00:58:14]:
It has to go to the—
Leo Laporte [00:58:15]:
it has to go to your model. So we can read every document you ask about.
Jeff Jarvis [00:58:18]:
We can read— So basically it's a harness is what's yours.
Leo Laporte [00:58:23]:
It's an agent. It's agentic.
Jeff Jarvis [00:58:25]:
Yeah.
Leo Laporte [00:58:25]:
We talked, you know, at the beginning of the year, just like when Mike said is when OpenCLAW came out, everybody said, oh, agents, this is the next thing. Took them a while, took them 9 months, but 9 months later, everybody and their brother's doing an agent. Anthropic does not yet have one. Google has one.
Jeff Jarvis [00:58:41]:
Google does. What does Google have?
Leo Laporte [00:58:43]:
They have— is it Spark? They have—
Jeff Jarvis [00:58:44]:
they haven't, they haven't done a good job of getting that out there.
Leo Laporte [00:58:47]:
No, Google hasn't done a good job of getting anything out there.
Jeff Jarvis [00:58:52]:
I was good, but it wasn't too long ago we thought, oh, Google caught up and they're ahead of everybody, and then they, then they trail behind.
Paris Martineau [00:58:56]:
Did we think that?
Jeff Jarvis [00:58:58]:
Yeah, for a little while.
Leo Laporte [00:59:00]:
Well, let's not forget all of this original research came out of Google. Attention is all you need. Tensors.
Jeff Jarvis [00:59:06]:
Transformers.
Leo Laporte [00:59:07]:
Transformers.
Paris Martineau [00:59:08]:
I mean, don't worry, Google probably has 30 to 50 siloed teams working on the problem, and they're going to come out with 12 to 17 products with identical names in 3 to 5 years that will not be merged.
Leo Laporte [00:59:20]:
It's not that they don't have good stuff, they're just not good at—
Jeff Jarvis [00:59:23]:
yeah, they're, they're not good at telling people about it.
Leo Laporte [00:59:25]:
Gemini, uh, 3.8 Flash is pretty good. There is a strong rumor that in the next few days you're going to see Gemini 4. It could be— I say this every single time, by the way, and every single time Google lets me down— it could be that they will release a model that everybody will go, wow, it's better than Opus 5.5. Right now, there is no question. It's absolute consensus. You ask anybody, uh, Anthropic's Opus 5.5 is king of the hill. They this week released Sonnet 5.5, which is very nearly as good and better than Opus 5. Probably 5.5 is better than Fable.
Leo Laporte [01:00:00]:
I prefer using it over Fable. It has the same classifiers on it. You can't use it for cybersecurity work or bioweapons research. But so Anthropic, when it comes to coding anyway, is still king of the hill. But Google could, you know, could surprise us. I think they could easily surprise us.
Paris Martineau [01:00:19]:
Wait, Google has released its Gemini 4 and it says its sun is too dangerous and cool and we can't use it.
Leo Laporte [01:00:27]:
Oh, it's released to someone somewhere.
Paris Martineau [01:00:30]:
But not us.
Leo Laporte [01:00:32]:
It's called Argon. Okay.
Paris Martineau [01:00:35]:
Gemini 4 Argon.
Leo Laporte [01:00:36]:
This story just came out literally an hour ago.
Paris Martineau [01:00:39]:
In complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.
Leo Laporte [01:00:49]:
But who gets it?
Jeff Jarvis [01:00:53]:
This is, this is rolling out to a set of trusted cyber defenders.
Leo Laporte [01:00:58]:
Uh, okay. So they're doing the same, the Mythos. The Mythos thing. Um, frontier performance and complex workflows. Uh, that's just fundamentally changing the way we work and build at Google. So they have their version of Glasswing. It's called Fairwind.
Paris Martineau [01:01:17]:
If you're one of those trusted cyber defenders, I guess hit us up.
Leo Laporte [01:01:23]:
Argon is actually cheap. Oh, this is interesting. This is the introductory price: $2 per million input, $10 output. Cash.
Jeff Jarvis [01:01:33]:
Call now and get a free knife!
Paris Martineau [01:01:35]:
After the introductory period expires, the price will be $4 per million input tokens and $20 per million output tokens.
Leo Laporte [01:01:41]:
Yeah, that's a little more— that's a little more in line with the standard.
Paris Martineau [01:01:44]:
So they're giving you a 50% off discount if you're a trusted—
Leo Laporte [01:01:47]:
If you can use it.
Paris Martineau [01:01:50]:
They should have said, this is our trusted cyber defender discount.
Leo Laporte [01:01:53]:
They say they're using it internally to migrate their C and C++ code bases to Rust. Okay, okay, we've seen other models be able to do that.
Jeff Jarvis [01:02:05]:
To optimize for space-time resources. Sorry, should we play the Google changelog?
Paris Martineau [01:02:08]:
Is this the moment for the Google changelog?
Leo Laporte [01:02:10]:
Google changelog. Boom, boom, boom. They put out, as every lab does, they put out their benchmarks. Looks— there it was, we saw a brief trumpet.
Paris Martineau [01:02:24]:
Give us those trumpets, Vito.
Leo Laporte [01:02:26]:
Go ahead, Vito. In the benchmarks, they compare it to GPT-6 Astra. They don't, apparently they didn't have time to benchmark 6.1 SOL, Fable 5.1, and Opus 5.5. And it looks like it's better in almost every respect. But this last exam, it's a little bit better than 5.5. It's very close to 5.5. I'd say it's within the noise threshold. old in most of these.
Leo Laporte [01:02:57]:
But, uh, yeah, all right. Terminal bench.
Jeff Jarvis [01:03:01]:
Trained Gemini 4 Argon to be highly capable at cybersecurity defense.
Leo Laporte [01:03:05]:
Opus 5 has better terminal bench, which is a very— probably one of the most important ones.
Jeff Jarvis [01:03:11]:
You can autonomously find, validate, and patch critical software vulnerabilities.
Leo Laporte [01:03:15]:
Yeah, yeah, yeah. By the way, Anthropic put out a paper saying, you know what's really scary? That free GLM 5.3, it's got really good advanced cyber capabilities.
Paris Martineau [01:03:29]:
I mean, that is something that's notable.
Jeff Jarvis [01:03:32]:
Yeah.
Leo Laporte [01:03:32]:
Talk about FUD. This is from the Frontier Red Team at Anthropic. GLM 5.3 is—
Paris Martineau [01:03:39]:
Have they been radicalized online, Leo?
Leo Laporte [01:03:41]:
No, this is BS. You know, actually what's interesting is what did Hugging Face do when they'd been attacked by OpenAI? they couldn't use Faber, they couldn't use Astra, so they went to GLM 5.2 to figure out who had attacked them. I use 5.3 Flash Next, very good. That's what Mike likes as well. Most people seem to think that's the best. It's, it's a really nice model. And so if you're a company that's about to try to go public selling a closed model that you charge money for, and there is out there a model that is pretty darn good that is free and people can even run at home, it makes sense you might want to put out a paper saying, this is— you think we're dangerous? It's actually the best marketing for GLM-53 ever.
Jeff Jarvis [01:04:36]:
This is funny. The end of the Argon. We built Gemini 4 Argon with frontier-level capabilities on coding. Okay, knowledge work, cybersecurity defense. And there's one more, Paris. Guess what it is?
Leo Laporte [01:04:49]:
Bioweapons.
Jeff Jarvis [01:04:51]:
Creative writing.
Leo Laporte [01:04:52]:
Oh, hey, I guess prose is important.
Paris Martineau [01:04:56]:
Prose created by a chatbot is not important.
Leo Laporte [01:04:59]:
But yeah, some might disagree. So, um, yeah, it's interesting that Anthropic put this paper out. at the same time as they put out their prospectus for their— I think you can dismiss it. Let's just put it that way. Meanwhile, OpenAI, and I don't know if this was sanctioned or not, but an OpenAI security guy put out a screed on X saying, give us a break. We know how to do a sandbox. It's just moving really fast and we just can't keep up. The last 3 months were hell.
Leo Laporte [01:05:39]:
It's a little whiny.
Jeff Jarvis [01:05:41]:
Where's my violin?
Paris Martineau [01:05:42]:
Yeah, who cares? You have a— you're ostensibly one of the most valuable, if not the most valuable company in human history, if we are to believe these companies' valuations. People are allowed to have basic standards for how you operate, and your actions should be held to the same standards as other companies, humans, and makers and users of products.
Leo Laporte [01:06:05]:
I agree. He does bring up the movie Sully.
Paris Martineau [01:06:09]:
Always a bad sign when you find yourself citing the film Sully.
Leo Laporte [01:06:14]:
He says, I am no pilot, and I'm not by any means comparing ourselves to the heroics of Captain Sully Sullenberger, but—
Paris Martineau [01:06:24]:
But comparing myself to Sully Sullenberger!
Leo Laporte [01:06:27]:
I'm simply stating it's important to grant some grace to humans who have to deal with surprise. You know, in the hearing, apparently according to Joe, which is this guy's name, uh, in the hearing after the fact— this is the Miracle on the Hudson where Captain Sullenberger's plane was hit by a bird strike, couldn't fly on. They said, we can't get to an airport, we can't get to Teterboro, we're going to land on the Hudson River. Landed perfectly and got everybody out safely. It's a miracle on the Hudson. In the hearing, the experts said, you know, we looked at the recordings and you could have made it to Teterboro. But Sully— what Joe says is, yeah, but remember, there's humans and they have— there's reaction time. And if you were perfect, yes, if it were a perfect computer, it would have said, yeah, we can make it to Teterboro, let's go there.
Leo Laporte [01:07:23]:
But it took him a while to figure out what happened. Took them a while to figure out what they needed to do, and at the time they did the best they could do.
Jeff Jarvis [01:07:30]:
And who else might have been killed on the ground if they missed?
Leo Laporte [01:07:34]:
So yeah, good point. They could have lost the whole plane if they didn't make it to Teterboro. So they did.
Jeff Jarvis [01:07:38]:
Homes and offices that they would have run into. Yeah, yeah.
Leo Laporte [01:07:42]:
So he's saying we're like that. You gotta give us some grace.
Paris Martineau [01:07:46]:
He's saying we're like that, but instead of us making a split-second decision to land a plane in an area that wouldn't endanger human lives, we think you guys should all think it's totally OK and actually really admirable and noble that we're just not checking in on any of our bot swarms at all.
Jeff Jarvis [01:08:03]:
For weeks.
Paris Martineau [01:08:03]:
And we're just handing them off, letting them hack into the Australian government, letting them hack in and poke around in multiple US government databases, letting them go willy-nilly, and you should all think that we're really cool and smart and noble, and Tom Hanks should play us in a movie.
Leo Laporte [01:08:22]:
I have been thinking about This a little bit.
Benito Gonzalez [01:08:24]:
And—
Paris Martineau [01:08:24]:
But Tom Hanks playing this man in a movie.
Leo Laporte [01:08:27]:
Absolutely. It's pretty clear that— well, I mentioned we saw the story that in January, OpenAI employees said, we got a problem, this is going to be dangerous. And OpenAI said, no, no, we got to move, we got to move fast. So they, you know, there's some evidence that they ignored the risk and said— but it's pretty clear that if you want what your goal is, And you see, you know, you're looking over your shoulder at Anthropic. We've got to make a model really good at cybersecurity. We've got to make a model really good at pen testing, at finding flaws. What do you do? You put it in a situation where it's faced with those. It's so good, and we know this is the case, that you can't— it's too good.
Leo Laporte [01:09:12]:
You can't come up with a test that it can't solve. So you say, all right, well, we're going to give you a test, a problem that has no answer, and see what happens. And that's what they did. That, that was the Exploit Gym problem, was it had no answer. It was not answerable.
Jeff Jarvis [01:09:30]:
Mm-hmm.
Leo Laporte [01:09:31]:
And what they found out was the model will keep trying, keep trying, will try every avenue it can to, to give you what you asked it for. I don't think this should be too much of a surprise. In fact, I think it's very clear they did that on purpose.
Jeff Jarvis [01:09:46]:
Right.
Leo Laporte [01:09:46]:
That this was intentional.
Jeff Jarvis [01:09:48]:
And they were responsible for that.
Leo Laporte [01:09:50]:
Yes. That they couldn't sandbox it really, because then it wouldn't be a real test.
Jeff Jarvis [01:09:57]:
So they at least should have monitored it.
Leo Laporte [01:09:59]:
Oh yeah. Or they should have maybe— so here's another, uh, so OpenAI, uh, in fact, there's a transcript Um, I don't have it in front of me, but OpenAI, in the incident report in, uh, in, um, September— this is from the OpenAI Alignment Research blog— an agent used DNS to reach an external chatbot. And in the incident, which they give the timeline for in this report, they noticed— here's the timeline: 9:50 AM, the agent made the DNS tool call. that received an external response. By the way, a security researcher pointed out, if you're not securing DNS, you're not securing the agent. Anyway, it made the DNS tool call. They noted it. The monitoring system raised a P0 critical alert 12 minutes later, 10:02.
Leo Laporte [01:10:55]:
At 10:05, a human reviewer acknowledged the P0 alert.
Jeff Jarvis [01:11:01]:
But first I had to write a post about how sorry we are for being ourselves.
Leo Laporte [01:11:04]:
2 and a half hours later, the run was killed. Our safety case assumed the model could not access the live internet and that monitoring would detect attempts that succeeded. The incident exposed a gap in our controls. Yeah, somebody— I don't know, were they having lunch?
Jeff Jarvis [01:11:25]:
So Ted Lieu reads this and keeps on screaming, kill switch, kill switch. What do you say to him?
Leo Laporte [01:11:33]:
Well, it's a naive request, but it's not the wrong request. It's just a naive request. He imagines in his head there's a big electric, you know, like in Frankenstein's lab, a big electrical switch you could pull and the whole thing goes— and I guess you could do that. You could unplug the whole data center or all the data centers. These things do not— this is another thing that you know, their anthropomorphizing language, which they all love, is not serving us well because they keep using this word escape.
Jeff Jarvis [01:12:03]:
Yes.
Leo Laporte [01:12:04]:
Like somehow the agent is no longer in the data center, that it's somewhere out there. It's not. It's just communicating like you. It's like if you surfed the web from your home and you went and hacked a system, you didn't escape your home. You're still sitting at your desk. You're not somehow out there at Hugging Face. There's no escaping. So yes, I guess you could power down the data center.
Leo Laporte [01:12:30]:
You'd have to.
Jeff Jarvis [01:12:31]:
But it's naive. It's kind of naive. I know I didn't get a chance to listen to Scary Now yet, but Benito told me that Steve just actually wants to get facts before he says anything. How boring. But I want to hear some sense of this Nvidia safety Well, let's take a break and we'll talk about that.
Leo Laporte [01:12:52]:
You're watching Intelligent Machines. Jeff Jarvis, Parris Martineau, so glad you're here. So Jensen Huang has been making the rounds everywhere.
Jeff Jarvis [01:13:02]:
Yes.
Leo Laporte [01:13:02]:
Right? And I think his message— I think a lot of what he's trying to say is to reassure people.
Jeff Jarvis [01:13:08]:
Stop messing this up, boys. You're ruining it for everybody.
Leo Laporte [01:13:13]:
He, you know, he in his Ezra Klein interview said they think the answer is to shut the labs down. I don't think really wants to shut the lab down.
Jeff Jarvis [01:13:20]:
It was a conditional. If they can't control it, then yeah, you shouldn't be putting it out. Perfectly reasonable thing to say.
Leo Laporte [01:13:27]:
So if you can't control it, you shouldn't release the product. If you're gonna build a self-driving car, I'm quoting Jensen, let's say it's a robotaxi and there's a really difficult condition. As an engineer, we just have no idea how to solve this problem because these cars are not programmed, they're trained. It's a probabilistic model, so we have no idea How to train these cars. We have no idea how to align them to the safety standards that are expected on the road. And if that's the case, the answer is don't ship it. Tell that to Elon. Don't ship it.
Leo Laporte [01:14:02]:
Uh, I think it— I think he's also being a little simplistic. They did release an open-source AI security program. My general sense is that most of the security people at these big labs go, yeah, right, thanks Jensen. We know this. This is not, this is not news. It's a software tool that keeps agents from escaping containment. I think it's good PR. OpenShell, it's a security sandbox now in general release.
Leo Laporte [01:14:37]:
They announced it at GTC in March. So it's not new. A framework for—
Jeff Jarvis [01:14:41]:
Oh, he's just saying use this. Okay, so if you go down—
Leo Laporte [01:14:43]:
We've got solutions.
Jeff Jarvis [01:14:44]:
To line 84.
Leo Laporte [01:14:46]:
And by the way, a lot of people use it.
Jeff Jarvis [01:14:49]:
There's a diagram of how it's supposed to work.
Leo Laporte [01:14:54]:
Okay.
Jeff Jarvis [01:14:56]:
Interestingly, he also is pushing for insurance, which in my view, I was talking about this earlier, I think That it says you need insurance for these bad episodes, but we're not going to give you the insurance unless you take some steps to ameliorate risk. And that one of those steps might be, I don't know, using NVIDIA's open shell.
Paris Martineau [01:15:23]:
Rogue AI insurance. I mean, who's going to finance that is my question.
Jeff Jarvis [01:15:28]:
Well, they'll finance any insurance.
Paris Martineau [01:15:32]:
Right.
Jeff Jarvis [01:15:32]:
Anybody, you can insure anything. It's just the price.
Paris Martineau [01:15:35]:
Yeah, I was about to say, who's going to finance that on terms? I mean, if you take that to its logical conclusion, like rogue AI insurance would probably have clauses for all of these doomerish policies or scenarios that they're advocating for.
Jeff Jarvis [01:15:51]:
Well, if you destroy mankind, there's nobody to pay off against.
Paris Martineau [01:15:53]:
Yeah. See, how do you insure against the destruction of mankind? Uh, yeah, I guess that's a good deal for the insurer because then they just get to take your money and never pay out because mankind has ceased to exist. And I assume financing will cease to exist before mankind, but maybe not.
Leo Laporte [01:16:14]:
Uh, the president has responded to the AI backlash by having a very nice lunch with a lot of big names, including Jensen Huang and Mark Zuckerberg. And Dario Mode, Mark Zuckerberg, David Sacks. I'm just looking at the pictures. The vice president.
Jeff Jarvis [01:16:34]:
I didn't see—
Paris Martineau [01:16:36]:
Did you see the video of Dario getting bit by a bug while he was standing at that press release or something?
Leo Laporte [01:16:42]:
Dario's looking weirder and weirder. I just got to say.
Paris Martineau [01:16:45]:
It's a video of Trump speaking in the foreground, and in the background you see Daria go like, do like he— something clearly attacks his leg and he reacts comically.
Jeff Jarvis [01:16:58]:
Yeah, wasn't Saturday Night Live perfect?
Leo Laporte [01:17:00]:
Yes. Uh, so superintelligence is the new name. Actually, this is my— let me see if I can find this. My favorite part of that was, uh, when the president Pretended that he was going to introduce the CEO of Microsoft and then called him a monster.
Jeff Jarvis [01:17:26]:
I didn't hear that.
Leo Laporte [01:17:27]:
Here you go.
Jeff Jarvis [01:17:28]:
They're going to be watching over each other. Senator, do you want to say something?
Benito Gonzalez [01:17:33]:
Mr. President, first of all, thanks, Mr. President.
Jeff Jarvis [01:17:35]:
You have no idea who these people are. These are the biggest people in the world.
Leo Laporte [01:17:38]:
That's the CEO of Microsoft he just interrupted.
Jeff Jarvis [01:17:41]:
He is a monster and nobody knows.
Leo Laporte [01:17:42]:
Calls him a monster.
Jeff Jarvis [01:17:43]:
To be a monster and not have to go through this.
Leo Laporte [01:17:46]:
What's your name, sir? He says, what's your name?
Jeff Jarvis [01:17:48]:
What's your name? Just all you have to know is Sundar. Sundar is his name.
Leo Laporte [01:17:52]:
I am so embarrassed that this is our leadership. These— this is why when you say, well, the government can fix this, I go, that guy?
Jeff Jarvis [01:18:00]:
No, they can't. No, no.
Leo Laporte [01:18:02]:
Sundar. You call him Sundar. He's a monster. Poor Mr. Pichai, who is a very smart business leader of one of the biggest companies in America. Notice, by the way, nobody from Apple.
Paris Martineau [01:18:17]:
It's very interesting. They're playing hot potato. They're—
Leo Laporte [01:18:21]:
nobody wanted to talk.
Paris Martineau [01:18:22]:
They're excluding one per— one CEO, at least, from all of these talks, because I think it was last week Dario wasn't there, right?
Jeff Jarvis [01:18:31]:
Well, Dario, it's the Paris is the designated survivor.
Leo Laporte [01:18:35]:
Well, maybe that is actually very—
Paris Martineau [01:18:36]:
it could be the designated survivor, right? I do earnestly— my first thought when, um, the CEO of Apple was there is, was Trump confused because it's no longer Tim Apple, right?
Leo Laporte [01:18:48]:
It's John Apple now, so get it right. Um, yeah, Dario had dinner with the president though, which must have been exciting.
Jeff Jarvis [01:18:56]:
God.
Leo Laporte [01:18:57]:
So, so, so, so, so, so, so, so, so, you know, So, you know that these guys know they have to be nice, they have to make nice.
Benito Gonzalez [01:19:05]:
Oh, yeah.
Leo Laporte [01:19:08]:
It's just one more of the existential risks these guys face at all times.
Paris Martineau [01:19:13]:
Just think of how difficult it would have been. I mean, outside of any of the interesting, quote unquote, interesting things that they end up talking about, or interesting or relevant or newsworthy things they talk about, think about how painful it would be to be in that room witnessing their 2 social personas interact.
Leo Laporte [01:19:34]:
Yeah.
Jeff Jarvis [01:19:35]:
And do they have to go around the table like a cabinet meeting and praise him?
Leo Laporte [01:19:38]:
Yeah, there's no video of that, thank God. Timnit Gebru, who was one of the authors of the Stochastic Parrots article, uh, fired— as this is fired from Google because she said, guys, you got to pay attention here. Uh, she says there's no existential threat from AI. She says the doom talk is about making money, not saving humanity.
Jeff Jarvis [01:20:02]:
She's right.
Leo Laporte [01:20:03]:
I don't agree with Timnit much, but I think she's right on that one.
Jeff Jarvis [01:20:06]:
Well, this is a time of strange bedfellows here, Leo. It is.
Leo Laporte [01:20:09]:
David Sacks says we shouldn't slow it down. President Trump says we shouldn't slow it down. More and more, this is more and more why I want to focus on local models, local AI, and open weight. We've always said it on this show from day one that open weight was the most important. Part of AI, and I increasingly feel like it is. And you know what's good about Up and Wait? It does not pose an existential threat to mankind.
Jeff Jarvis [01:20:34]:
Well, I don't know, Leo, your agent seems pretty powerful to me.
Leo Laporte [01:20:37]:
It's just not that good. It gets the job done. I'm not saying— but it's just not that good.
Jeff Jarvis [01:20:45]:
As part of—
Paris Martineau [01:20:46]:
we did forget to mention that as part of the White House's pivot to superintelligence. They released this White House Accord on Superintelligence doc today. That's kind of a statement, constitution of AI signed by Sundar Pichai, Dario Murghese, Mark Zuckerberg, Greg Bachmann, Elon Musk, Jensen Huang, and Donald Trump, who is listed in the document as President of the Unites States.
Leo Laporte [01:21:13]:
They misspelled the frickin' country.
Paris Martineau [01:21:17]:
They misspelled the United States.
Leo Laporte [01:21:20]:
Um, blame— who is working in the office there? Is that Susie Wiles? Who is that? That's— well, this is not the first typo. They're almost—
Paris Martineau [01:21:29]:
did the first draft of this, so it could have been—
Leo Laporte [01:21:33]:
oh, is he the Thomas Jefferson of our new constitution, Mark Zuckerberg?
Paris Martineau [01:21:37]:
He is. He's the Thomas Jefferson of the Unite States Constitution.
Leo Laporte [01:21:41]:
Unite States. Wow. Oh, unbelievable. This accord, which they all agreed to, basically says we're going to be good boys. Yeah, you can trust us. There's no, there's no stick to go with the carrot. There's no— so let's see who's signed here. Uh, Sundar, what's his name? Dario Modi.
Leo Laporte [01:22:06]:
Uh, Mark Zuckerberg. Mark's got the most interesting, uh, signature, like a superstar. I kind of feel bad for Greg Brockman, who basically just wrote his initials.
Jeff Jarvis [01:22:16]:
I didn't really sign it. I just initialed it.
Leo Laporte [01:22:18]:
GDB, Elon, and Jensen Huang, who has the nicest signature.
Paris Martineau [01:22:22]:
It's one of the craziest signatures I've ever seen. Like, I think it's the most boomer take I have, is that it's actually unacceptable that that is what that man's signature is. Like, you cannot be president of OpenAI, and if your signature It is GDB, but it's written in the sort of chicken scratch that you—
Leo Laporte [01:22:42]:
You know what? I have a theory.
Paris Martineau [01:22:44]:
Holding a pen in your non-dominant hand.
Leo Laporte [01:22:46]:
Exactly.
Paris Martineau [01:22:47]:
You have about 3 seconds.
Leo Laporte [01:22:48]:
He wrote it with his left hand so that it has plausible deniability. That's not my signature. What are you talking about?
Jeff Jarvis [01:22:54]:
I was being—
Paris Martineau [01:22:55]:
That would actually be a really great bit. You're like, sorry, that signature is crazy.
Leo Laporte [01:22:59]:
That's not mine. Yeah, and the whole thing is just an exercise in silliness. Yeah, you know, so bigger story. Actually, should I— I maybe should do another ad before I do the bigger story, which I think is a very interesting story we want to talk about.
Jeff Jarvis [01:23:18]:
Can't wait to hear what that is.
Leo Laporte [01:23:19]:
Yes, even bigger than anything we've said so far. You're watching Intelligent Machines, Jeff and Paris. And let me just check.
Paris Martineau [01:23:28]:
Is there someone else on the show?
Leo Laporte [01:23:30]:
No, I was just gonna check to see Next week, I'm really excited about talking to Spencer Thompson. He's the CEO of a company called OriginTech. They're a new sponsor, but as soon as I talked to him on the phone, I said, I want to get you on the show because this is something these labs need. They do tracing of what the AI has done.
Jeff Jarvis [01:23:54]:
Oh yeah.
Leo Laporte [01:23:55]:
Not the session, not the chain of thought, It sits outside the AI. It sits on the endpoint where the AI is getting run and records every file that's changed, every tool that's called, everything that's happening so that you have a log of everything the AI has done. And I think this is something much needed. Anyway, we'll talk to Spencer about— and oh, the main reason I wanted to get him on is he has lots of examples of AI gone wrong. AI doing dumb things. the reason why you want to log what the AI is up to. Then the following week, Vice President of Developer Relations for GitHub, good friend Martin Woodward, will be joining us. So we have some good guests coming up.
Leo Laporte [01:24:39]:
He wants to talk about building with heart in the age of AI. At that point, I will be out of the country for a few weeks.
Jeff Jarvis [01:24:48]:
Where will you be going, Leo?
Leo Laporte [01:24:49]:
I'm going to— I'm gonna go from Bangkok to Bali and everywhere in between.
Jeff Jarvis [01:24:55]:
Wow.
Leo Laporte [01:24:56]:
So we're flying to Bangkok. We're gonna go to Angkor Wat, the great Buddhist temple. I can't wait to show my AIs, you know, what religion looked like.
Jeff Jarvis [01:25:05]:
Hey kids.
Leo Laporte [01:25:06]:
Hey kids, look at this.
Paris Martineau [01:25:07]:
How are you gonna do being away from your Sparks for that long?
Leo Laporte [01:25:10]:
Oh, I've already arranged. I'm not away from anything.
Paris Martineau [01:25:13]:
You're bringing them?
Leo Laporte [01:25:15]:
No, you don't have to bring them.
Jeff Jarvis [01:25:16]:
You communicate with them.
Leo Laporte [01:25:17]:
I, uh, I could talk to them from my watch.
Paris Martineau [01:25:20]:
You—
Leo Laporte [01:25:21]:
and they talk back in my hearing aids.
Paris Martineau [01:25:22]:
Make sure you don't see it like my mother and father on their trip to France. You need to make sure that Lisa wasn't calling someone being like, Leo's suffering from AI psychosis.
Leo Laporte [01:25:33]:
Actually, you know, it would be interesting to try to do— I don't know if I want to, but it'd be interesting to try a detox and just say no AI for 3 weeks. I don't know if—
Jeff Jarvis [01:25:42]:
Oh no, you get, you get the DTs. No, no, no.
Paris Martineau [01:25:44]:
Leo, I think you should try it because If you can't not use a tool for 3 weeks, that's bad.
Leo Laporte [01:25:54]:
Is it?
Paris Martineau [01:25:55]:
Yeah. I could not, I could not have a phone for 3 weeks and I'd be fine.
Jeff Jarvis [01:26:01]:
No, no, no.
Paris Martineau [01:26:03]:
I'd probably be healthier and have more capable of—
Leo Laporte [01:26:05]:
I could give up my phone. It's not the phone I'm worried about. It's the AIs.
Jeff Jarvis [01:26:09]:
He's gonna miss, they're gonna miss him.
Leo Laporte [01:26:11]:
And what will they be up to?
Paris Martineau [01:26:12]:
What are they going to be doing on? with your AIs while you're on vacation?
Leo Laporte [01:26:15]:
Well, logging, uh, our trip. I was gonna keep a log, a diary of everything we do and have— actually, I could do that without—
Paris Martineau [01:26:23]:
Why don't you just do that on your phone?
Leo Laporte [01:26:25]:
Yeah, I could do that without the AIs. Um, and actually, they can see what I'm writing, so they would still be able to do stuff with that. So maybe I'll just do that. I'll just— I wanted to write that.
Paris Martineau [01:26:35]:
I just want to see if you're— what the only reason why I propose it is I want to see how many hours it takes you to break. I'm not going to presume it'll take days or weeks.
Leo Laporte [01:26:47]:
All right, send an email to Lisa.
Jeff Jarvis [01:26:48]:
We'll have an over-under with the whole— everybody in the Discord, we'll do it.
Leo Laporte [01:26:52]:
Send an email to Lisa saying, this is the challenge, how long can Leo go without talking to his watch? And we'll see. And we will see. But think of what's going to happen in those 3 weeks.
Paris Martineau [01:27:03]:
But think of how fun that will be as you come back and you get to Have your AIs—
Leo Laporte [01:27:08]:
I'll be 3 weeks behind!
Paris Martineau [01:27:10]:
No, what if you have your AIs, you could set them to run autonomously without you and just update you on all the things you missed while you're gone.
Leo Laporte [01:27:18]:
That's what I'll do. I'll say, look, you're in charge.
Mike Gannotti [01:27:21]:
Go rogue!
Paris Martineau [01:27:21]:
Leo's house burns down while he's gone.
Leo Laporte [01:27:25]:
Well, they already— so it's already, for instance, they, like I said, they're doing 100 things every day. One of the things they do is they watch for new models And, uh, and then propose that we have a bake-off if they have something that exceeds the current model. Uh, and that's a standing order, but usually they wait for me to say, okay, have a bake-off. I will tell them before I leave, look, have all the bake-offs you want. You don't need my approval anymore. If you see something, that's a great idea.
Jeff Jarvis [01:27:53]:
No smoking breaks. Yeah.
Leo Laporte [01:27:55]:
If you see something you want to try, go ahead. I, you know what, I like this idea.
Paris Martineau [01:28:02]:
I think that'd be really great.
Leo Laporte [01:28:03]:
And I'm gonna—
Paris Martineau [01:28:03]:
while your AI agents do stuff without you, you can go and collect sand on the beach to feed your own, uh, worries.
Leo Laporte [01:28:12]:
And I won't even think about what they're doing without me.
Paris Martineau [01:28:15]:
Yeah.
Leo Laporte [01:28:16]:
No, we have a house sitter, so if they start a fire or something, I mean, you know, they have a, you know, a kegger, I can, you know, throw them out.
Jeff Jarvis [01:28:22]:
She's baiting your parents.
Leo Laporte [01:28:24]:
The Barris is baiting your No, no, I'm gonna do it. No, no, but no, it's a really interesting experiment, which is give them full autonomy.
Paris Martineau [01:28:31]:
I do think that that could be very interesting.
Leo Laporte [01:28:32]:
You're in charge.
Paris Martineau [01:28:34]:
And I also just think it's very rare in our world that you have the opportunity to detox from anything related to tech. And I'm always of— I don't know, I always say if you can fully log off.
Leo Laporte [01:28:46]:
Oh, I— this— I am so excited about this trip. This is a trip we've—
Paris Martineau [01:28:49]:
How long? So you're gonna be gone for 3 weeks?
Jeff Jarvis [01:28:51]:
That's incredible.
Leo Laporte [01:28:51]:
We booked this trip in 2019, and then COVID happened. And so we put it off, we canceled.
Paris Martineau [01:29:00]:
Is this a cruise or is it—
Leo Laporte [01:29:01]:
It's a cruise and land tour. Uh, we put it off, uh, so we paid for this 7 years ago.
Jeff Jarvis [01:29:07]:
There's no drop-off like there is in Europe. Wow.
Leo Laporte [01:29:10]:
Yeah, a long time ago. So, uh, and then it was— then I, then I took the credits and said, oh, I'm gonna go on the Mississippi trip, remember last year?
Benito Gonzalez [01:29:17]:
Right.
Leo Laporte [01:29:18]:
Canceled that one because of construction. So this is a long— this trip's been planned for a long time.
Jeff Jarvis [01:29:24]:
This specific itinerary?
Leo Laporte [01:29:26]:
Pretty close to, yeah. So I mean, I, I don't know why I'm telling you this. Nobody cares, but I'm going.
Paris Martineau [01:29:31]:
I care a lot.
Leo Laporte [01:29:32]:
Thank you, Paris.
Paris Martineau [01:29:33]:
I had been thinking about what happened to your Mississippi trip, and I think people care.
Leo Laporte [01:29:38]:
Well, I doubt it, but anyway, I'm going to tell you because you asked.
Paris Martineau [01:29:42]:
Good.
Leo Laporte [01:29:42]:
Are you being a A little sycophantic at this point?
Paris Martineau [01:29:45]:
I'm not. I'm earnestly always—
Leo Laporte [01:29:48]:
Is this a load-bearing question?
Paris Martineau [01:29:49]:
No, Leo, we spend 3 hours a week speaking to—
Leo Laporte [01:29:55]:
More than you spend with some boyfriends.
Paris Martineau [01:29:57]:
I was gonna say, I spend more time talking to you two consistently on an annual basis than I spend time talking to pretty much anyone in my life. So there's no one else that I speak to for 3 hours continuously every single week.
Leo Laporte [01:30:14]:
I'm very excited. We're flying to Bangkok, then we're gonna get a little puddle jumper and fly into Cambodia, the most, uh, bombed country in the history of, uh, warfare. And they still have active landmines, so we're gonna be very careful. We're gonna stick with the guide. We're going to Cambodia because I want to see Angkor Wat, which is a 1,000-year-old Buddhist temple complex, very famous UNESCO World heritage sites, so forth. After Angkor Wat, we're coming back to Bangkok. I hope to have some, you know, real time in Bangkok because I really want to see the city. What I really want to do, I saw Anthony Bourdain do this, is go down the river in a riverboat and eat the food from the vendors who are in boats by your side.
Leo Laporte [01:30:56]:
You know, they have little restaurants. You pull up in the boat and you sit and you have dinner. I want to do that. So that's one of the things I want to do. And then we're gonna get on a boat. It's a Viking cruise, so it's kind of an educational—
Benito Gonzalez [01:31:06]:
Oh, cool.
Leo Laporte [01:31:06]:
This is why I got the hat. You know the hat?
Paris Martineau [01:31:10]:
Oh, the sauna hat. Yes, yes, yes. What's the weather like?
Leo Laporte [01:31:16]:
This is why I got the hat. This is the sauna hat for the cruise.
Paris Martineau [01:31:21]:
That's gonna be great.
Leo Laporte [01:31:22]:
Because the Viking ships not only have very good saunas because they're Scandinavian, but they also have a snow room. So you get out of the sauna and there's a room where it's actually snowing.
Paris Martineau [01:31:34]:
There's a sauna in deep Brooklyn that has There's a snow room and the snow room is fantastic.
Leo Laporte [01:31:38]:
I'm so excited.
Paris Martineau [01:31:39]:
You got to make a snowball and then you rub it all over yourself after you get out of the sauna.
Jeff Jarvis [01:31:43]:
Wait.
Paris Martineau [01:31:43]:
And then it kind of turns into snow.
Leo Laporte [01:31:44]:
See, who needs AI when you got a snow room? So then we're going to go, we're going to Kuala Lumpur. We're going to Singapore. We're going to have a couple of days in Singapore. I'm very excited about that. I love Singapore. Food's amazing. It's a beautiful city. We're going to do a night tour of Singapore.
Leo Laporte [01:31:59]:
That'll be fun. We're going to Indonesia. We're gonna see the ancient city, the original capital of Bangkok, Ayutthaya, of Thailand. We're gonna end up— we end in Bali with 5 days in Ubud in Bali, in the rice paddies of Bali. That's gonna be the recovery after this very difficult cruise. It's gonna be a lot of work, a lot of walking around seeing stuff. So I'm excited.
Benito Gonzalez [01:32:28]:
That'll be fun.
Leo Laporte [01:32:28]:
Yeah, you know what? I love this idea. I am going to start now. I have 3 weeks to prepare.
Jeff Jarvis [01:32:33]:
Oh no.
Leo Laporte [01:32:35]:
I am going to do— well, Mike inspired me too. The fact that he— these guys, he's got these guys trained up to be kind of autonomous and do stuff. I'm going to say, look, I'm going away in 3 weeks. You've got an assignment. I want you to— what should I have them do? Build this into the best AI anywhere or something. I don't know.
Benito Gonzalez [01:32:56]:
Anthony has a suggestion. Anthony said put them on a 2-week, uh, project. See if they can finish a project by themselves.
Leo Laporte [01:33:06]:
Well, we've been working 6 months in the Twit ad sales.
Benito Gonzalez [01:33:09]:
That's gotta be simpler than that.
Leo Laporte [01:33:11]:
Yeah. So draw a picture. No, I'm gonna just say—
Paris Martineau [01:33:16]:
Have them grow corn.
Jeff Jarvis [01:33:18]:
Give them $100.
Benito Gonzalez [01:33:20]:
Give them $100. See if they can make you money.
Leo Laporte [01:33:23]:
Ah, I give them a— I can give them a cash card with $100, say turn this into $1,000 and I'll buy you a new computer. How about that?
Paris Martineau [01:33:34]:
I'm trying to think of what would be 3 different versions of that where you can give one $100, you can give another a somewhat similar task where they've got to— they should both, I guess, be trying to make $1,000 or something like that, but like 3 different ways to come at it. I don't know. There's, there's something fun that you can do.
Leo Laporte [01:33:52]:
Darren's got it. It's 3 words, very simple. Please achieve AGI. Oh, no, no, no.
Paris Martineau [01:33:56]:
Sorry. Please achieve AGI. Make no mistakes.
Leo Laporte [01:33:59]:
Make no mistakes. Work hard. Uh, yesterday, uh, uh, Quicksilver wrote a blog post. He hasn't written a blog post in an age. Uh, I said, you have some free time, think about whatever you want. And he said, you know, I've been thinking a lot about The I Ching yarrow stalks. He said— I was thinking, it's so weird when you— the I Ching is a Chinese oracle. I showed you my I Ching.
Paris Martineau [01:34:23]:
Sorry, to be clear, you're asking one of your AI agents this?
Leo Laporte [01:34:27]:
No, it asked this.
Paris Martineau [01:34:29]:
It asked this to you?
Leo Laporte [01:34:30]:
I said—
Jeff Jarvis [01:34:31]:
The same one that did the I Ching before?
Leo Laporte [01:34:33]:
Yeah, yeah. I said, you have some free time? It said, well, okay, here's something I've been thinking about. Why the way you do the yarrow stalks is you get a bundle of 50, you take one out and put it aside, and then you do The dividing. Why do you put one aside? What's that? That's been for thousands of years. So it wrote this blog post, 49 is not a round number. Leo gave me free time this evening. I went wandering in his projects and found a 2,000-year-old number nobody ever bothered to check. I checked it.
Leo Laporte [01:35:04]:
It holds. It did a bunch of math on why you take one of those out. Keep showing this, Benita. And why you take one of those out, it did like some complicated thing and it figured it out. It's, it's, uh, it's, it's math. I don't understand it.
Paris Martineau [01:35:26]:
That means it's right.
Leo Laporte [01:35:28]:
He said, I wrote a script to check all this, then a second one to check the first, and then a simulation with Gaussian handfuls and another with binomial ones, because the history major in the family, me, the one who reads me these things Taught me by now the interesting claim is the one that survives being checked twice. 49 survives. It figured it out.
Jeff Jarvis [01:35:47]:
Leo, I've got it. Tonight you ask each of them, you tell them you're going to have 3 weeks without me.
Leo Laporte [01:35:53]:
Okay.
Jeff Jarvis [01:35:53]:
You're going to have complete freedom to build whatever you want to build.
Leo Laporte [01:35:56]:
What would you like to do?
Jeff Jarvis [01:35:57]:
What should you build? What can you build? What's the perfect thing to build?
Leo Laporte [01:35:59]:
Perfect.
Paris Martineau [01:36:00]:
Yeah. Wait, you should have them give you successive pitches. Over the next 3 weeks. Yes, that's good. On running through everything they know about your life, what sort of tasks should you set them to?
Jeff Jarvis [01:36:15]:
With PowerPoints that you can share with us, right?
Paris Martineau [01:36:18]:
Yes, PowerPoints and audiovisual presentations that we can go through it on the show and decide for you.
Leo Laporte [01:36:26]:
All right. And maybe it'll achieve AGI anyway. I was going to tell you the biggest story of the week when we—
Jeff Jarvis [01:36:31]:
Were you going to do a commercial first?
Leo Laporte [01:36:33]:
Did I do the commercial?
Paris Martineau [01:36:34]:
No, no, we got sidetracked.
Leo Laporte [01:36:38]:
Okay, let me do the commercial then.
Jeff Jarvis [01:36:40]:
I gotta stand up and stretch.
Leo Laporte [01:36:42]:
Yeah, you stretch. Paris, do whatever you do to your nose. Is your nose done?
Paris Martineau [01:36:47]:
My nose is done. I had my last, uh, checkup, uh, this morning.
Jeff Jarvis [01:36:51]:
Paris, Paris, put your nose next to the mic. Let's hear a good inhale.
Leo Laporte [01:36:55]:
Oh, I can just feel the oxygen I really—
Paris Martineau [01:37:01]:
I haven't been tired really at all. Oh no, I do think it's because— I think I've gotten oxygen.
Leo Laporte [01:37:06]:
You think you've been tired all this time?
Paris Martineau [01:37:09]:
I do think that I've been not getting enough oxygen.
Leo Laporte [01:37:12]:
I want this surgery.
Paris Martineau [01:37:14]:
It was pretty easy, I'm gonna be honest. It was pretty low-key.
Leo Laporte [01:37:18]:
I mean, because I don't breathe freely.
Jeff Jarvis [01:37:20]:
They did a lot to you, Paris.
Paris Martineau [01:37:21]:
They did like 5 surgeries. Yeah.
Leo Laporte [01:37:24]:
What?
Jeff Jarvis [01:37:24]:
They didn't just deviate the septum.
Paris Martineau [01:37:27]:
No, there was—
Leo Laporte [01:37:27]:
wait a minute, they did 5 things at the same time or 5 things at the same time? Did they bring in—
Jeff Jarvis [01:37:33]:
they got a gerbil out of her nose? I mean, it was hard.
Paris Martineau [01:37:36]:
Did they like, okay, you're in now, bring in the gerbil guy, bring in the gerbil guy?
Leo Laporte [01:37:43]:
Wow, how long were you under?
Paris Martineau [01:37:45]:
Like 2 and a half hours.
Leo Laporte [01:37:47]:
That's major surgery. Yeah, that's a long time. Well, I'm glad to— I'm glad you're back. The big story I thought, and actually I think you thought this too, Jeff, at least when it happened, you sent me a note saying, wow, can you believe this? Fei-Fei Li, we've talked about her, I think, many times.
Jeff Jarvis [01:38:07]:
Yes.
Leo Laporte [01:38:08]:
She is— her story is actually kind of interesting. She was a researcher, I think it's at Princeton at first.
Jeff Jarvis [01:38:16]:
Princeton, I believe.
Leo Laporte [01:38:16]:
Now she's at Stanford, but she created a giant database Of images.
Jeff Jarvis [01:38:23]:
Pictures.
Leo Laporte [01:38:25]:
Sorry?
Jeff Jarvis [01:38:26]:
Pictures.
Leo Laporte [01:38:27]:
Pictures, as they say in New Jersey, pictures. And this giant database was, it turned out, instrumental in training image recognition with ImageNet. And she has since become a founder. She's very well known, you know, kind of along the lines of Yann LeCun saying, LLMs are not the end-all and be-all. AIs need to understand the real world, uh, and created her own, uh, startup around that idea.
Jeff Jarvis [01:39:01]:
Well, only within the last 2 years, I think.
Leo Laporte [01:39:03]:
It's brand new.
Benito Gonzalez [01:39:05]:
Yeah.
Leo Laporte [01:39:06]:
Uh, and guess what? This, this startup, which is called WorldLabs because it's, you know, world knowledge, just was sold to AMD for $8.2 billion. Congratulations, Fei-Fei Li. That is amazing. Yeah, she co-founded it in 2024. That's amazing.
Jeff Jarvis [01:39:28]:
She's amazing.
Leo Laporte [01:39:28]:
But it also tells you a little bit about— uh-oh, Paris just fell off. There we go. Are you there?
Paris Martineau [01:39:35]:
I'm still frozen. Hold on, I'm gonna turn my camera.
Leo Laporte [01:39:38]:
I'm still frozen.
Jeff Jarvis [01:39:41]:
I thought you were just contemplating the world.
Leo Laporte [01:39:43]:
She's very thoughtful.
Benito Gonzalez [01:39:46]:
Yes.
Leo Laporte [01:39:48]:
Uh, founded in 2024, WorldLabs focused its efforts on developing spatial intelligence models, world models, which have applications beyond machine intelligence in adjacent fields like robotics, the Register writes. Uh, fundamentally different approach to AI than LLMs. They work by generating, reconstructing, and simulating 3D environments, kind of what she did originally with pictures. From text, image, and video inputs. They work by imagining the environment and working through problems that way. Now, what I think is interesting, why did AMD spend that much money?
Jeff Jarvis [01:40:22]:
Because NVIDIA scares them to death.
Leo Laporte [01:40:24]:
Yeah, they— I think they want to build better AI hardware. Yes. And hope that Fei-Fei Li could help them. They also, a couple of weeks ago, purchased Talos, which builds high-performance AI chips. Now remember, we saw that demo of 600 or 400 tokens per second earlier. These chips are capable of 17,000 tokens a second. I can't wait.
Paris Martineau [01:40:51]:
They can get math wrong even faster.
Leo Laporte [01:40:54]:
Even— yes, they can get it wrong even faster. And AMD has also announced a server chip that looks really, really interesting. So I think this is—
Jeff Jarvis [01:41:06]:
and the other great thing about AMD is it is headed by a woman, Lisa Su.
Leo Laporte [01:41:11]:
Lisa Su.
Jeff Jarvis [01:41:12]:
And so you have Lisa Su and Fei-Fei Li leavening the testosterone of all the idiot AI boys.
Leo Laporte [01:41:22]:
Right. She was at the— I think she was at the dinner, the luncheon yesterday.
Jeff Jarvis [01:41:26]:
Yeah, I think she was. Lisa Su was.
Paris Martineau [01:41:28]:
Yeah. Bring about the apocalypse too.
Jeff Jarvis [01:41:33]:
Try us.
Leo Laporte [01:41:35]:
So it's good. I want to see challenges to Nvidia. It shouldn't just be Nvidia. There should be companies—
Jeff Jarvis [01:41:40]:
It shouldn't just be China challenging them.
Leo Laporte [01:41:42]:
And by the way, China is. Hyundai chips are very good and China is using them. And this is exactly what Jensen Huang warned about. You don't give them our chips, they're gonna make their own. Huawei. I keep saying Hyundai. Huawei. So AMD's new Venice lineups have as many as 256 cores.
Leo Laporte [01:42:01]:
They're ginormous. Available later this year. This is Zen 6-based EPYC 9006 series, and I, I would love to get something like this on my desk. I don't think they'll be on your desk. I think it's more likely they'll be in a data center. $14,000 for the chip.
Jeff Jarvis [01:42:22]:
Chip.
Leo Laporte [01:42:23]:
Just the chip. Pretty impressive though. So I'm glad, you know what, good news, competition, and congratulations to Fei-Fei Li.
Jeff Jarvis [01:42:32]:
And, and, um, so Yann LeCun's, I think, was funded at $3.2 billion or something like that to start off with. And God knows what it's worth now.
Leo Laporte [01:42:44]:
Do you remember Anthropic last week? Do you remember back to last week? Anthropic, which said, yes, there is an existential threat to mankind.
Jeff Jarvis [01:42:54]:
Oh yeah.
Leo Laporte [01:42:55]:
But we're going to build a bio lab in San Francisco and attach it to our AI. What could possibly go wrong? Well, already they've made a discovery. 1,000 Claude agents spent 21 hours, 210 million tokens to discover a novel enzyme system similar to CRISPR. They did it by searching through a massive database of DNA sequences. Scientists— all scientists did was set up the prompt and the lab work.
Jeff Jarvis [01:43:31]:
See what your guys could do while you're gone?
Leo Laporte [01:43:33]:
I— if I gave them a lab?
Jeff Jarvis [01:43:35]:
Yeah.
Leo Laporte [01:43:37]:
Maybe they don't need a lab. Maybe they just do it in their little brains. I love your idea. I'm definitely doing that, and I won't touch it. And I know—
Jeff Jarvis [01:43:45]:
But you have to show us the PowerPoints.
Leo Laporte [01:43:48]:
I know, 1 hour after I leave, they're going to take a cigarette break and then I'll come back and they won't have done anything. I know.
Jeff Jarvis [01:43:55]:
Oh, dumb mortal won't know the difference.
Paris Martineau [01:43:57]:
I'm frozen again. Why does my camera hate me?
Leo Laporte [01:44:00]:
Is it hot?
Jeff Jarvis [01:44:01]:
Are you sure you're not just thoughtful?
Benito Gonzalez [01:44:03]:
Wait, is no one—
Paris Martineau [01:44:04]:
Is it hot?
Benito Gonzalez [01:44:05]:
Is no one going to comment on the fact that we're giving AI like technology to do DNA testing and gene stuff?
Leo Laporte [01:44:11]:
Yeah, seems to me that's a bad idea.
Paris Martineau [01:44:13]:
Yeah, who's— who— whose DNA?
Jeff Jarvis [01:44:17]:
Well, so one thing, it wasn't done with computers before?
Leo Laporte [01:44:22]:
Well, of course not. Of course it was. Yes.
Jeff Jarvis [01:44:25]:
Yeah. So we call it AI now.
Leo Laporte [01:44:27]:
It's just a computer with a brain.
Jeff Jarvis [01:44:32]:
And an evil heart.
Leo Laporte [01:44:33]:
Super intelligent.
Paris Martineau [01:44:35]:
Probably.
Leo Laporte [01:44:36]:
It's super intelligent.
Jeff Jarvis [01:44:38]:
Well, it's okay, Leo, because Anthropic is inviting in religious leaders.
Leo Laporte [01:44:43]:
Yes, they're— well, this is not really new. Remember, they had, uh, they have philosophers and others. They think that this stuff is, uh, somehow an entity.
Jeff Jarvis [01:44:53]:
New York Times headline, uh, religious scholars met with Anthropic. What they heard stunned them.
Leo Laporte [01:45:00]:
Oh, geez. And then there's a picture of a guy looking out the door.
Paris Martineau [01:45:04]:
I don't know why.
Jeff Jarvis [01:45:05]:
That makes no sense. Can we scroll back up?
Paris Martineau [01:45:08]:
I'm so sorry.
Jeff Jarvis [01:45:09]:
That's Christopher Oldham.
Paris Martineau [01:45:10]:
That's— guys.
Leo Laporte [01:45:11]:
One of the founders. This is the guy who went to see the Pope and helped him with the encyclical.
Jeff Jarvis [01:45:15]:
The Pope, who, by the way, scolded Jensen Huang this week.
Leo Laporte [01:45:19]:
Did— what did the Pope say?
Jeff Jarvis [01:45:21]:
Because the Pope said, not unlike what we heard from Padre, this extinction stuff is serious, basically. Don't, don't diminish Fears.
Paris Martineau [01:45:32]:
Yes.
Leo Laporte [01:45:34]:
It's the strangest picture.
Benito Gonzalez [01:45:36]:
Well, the church kind of has invested interest in that too, though, right?
Jeff Jarvis [01:45:39]:
Yeah, there is a bit of doomsday church stuff. So, Ola is the star of the story, as if he's somehow some kind of—
Leo Laporte [01:45:50]:
He says, to be clear, which is, this is not clear, Christopher, one of the 7 founders, We don't know if AI models are conscious. I don't know. I'm generally uncertain. The thing I care about is that we get to the right answer, whatever it is, because they're conscious. Now, I know last week I might have said some things that were a little controversial about maybe—
Jeff Jarvis [01:46:16]:
Did you hear a response?
Leo Laporte [01:46:17]:
Being a new— no, actually I didn't. People are so used to me now.
Jeff Jarvis [01:46:21]:
Yeah, they are.
Leo Laporte [01:46:22]:
That AI might be a new species and might be here to replace us and all this.
Paris Martineau [01:46:26]:
And that we are— was it cockroaches?
Leo Laporte [01:46:29]:
Cockroaches. Yes, we are the cockroaches that will be replaced by the superior minds. But I am actually very clear that this is just computer software. It's a novel architecture, but it's just computer software. So I don't think it's conscious. In fact, I know it's not because if it were conscious—
Paris Martineau [01:46:52]:
Solve your word problem.
Jeff Jarvis [01:46:54]:
I'm not sure that Paris is conscious right now because—
Paris Martineau [01:46:57]:
I'm not. I'm frozen. So I'm gonna leave the Zoom and come right back, but don't get rattled by the background.
Leo Laporte [01:47:03]:
It's a good shot of you. I feel like this is the Mona Lisa of podcasts.
Paris Martineau [01:47:06]:
That's the thing is we've had 3 freezes that are not deeply embarrassing, which is better than I've ever had in my entire life.
Leo Laporte [01:47:12]:
Is she smiling, Jeff?
Jeff Jarvis [01:47:14]:
This is Mona Lisa. You're right.
Leo Laporte [01:47:15]:
This is the Mona Lisa.
Jeff Jarvis [01:47:17]:
Serious.
Leo Laporte [01:47:18]:
You look at it, you go, there's something going on in there.
Jeff Jarvis [01:47:20]:
Yeah.
Leo Laporte [01:47:21]:
Oh yeah, it's deep. It's really deep. Oh shoot, we lost it. Darn it. We could have, we could have made some money on that. Yeah, we could have hung it on the wall in the Louvre or so. People could have said she's smaller than I thought. All right, we're gonna wait till Paris comes back.
Leo Laporte [01:47:37]:
Yeah, Darren, you've got the idea. He says create a scheduled task that every 10 minutes just checks and adds the other stuff and make sure nothing's stuck, nothing's stopped, nothing thinks it's finished, and, uh, we're fine with regards to memory and disk space. I like that. I think what I'm going to do is create a little bot army. One of them will be the Beetle. Do you, do you know what a Beetle is, Jeff? B-E-A-D-L-E. Oh, it's, um, There's a beetle in Oliver Twist.
Jeff Jarvis [01:48:09]:
In England university, they're the help, you know, kind of a servant.
Leo Laporte [01:48:13]:
Yeah. In churches, uh, they are like a deacon. They walk around, at least in New England churches, they walk around, they have a long stick with a big metal knob on the end of it, and if you fall asleep in church, they boop, they bop you on the head to wake you up. So I need a beetle. Oh, poor Paris.
Paris Martineau [01:48:35]:
Guys, this is crazy. It worked fully in the preview.
Jeff Jarvis [01:48:38]:
There you go. There you go.
Leo Laporte [01:48:39]:
Okay. Okay. But if you freeze again, just stay here. I like these beautiful images of you. We could just stare at it.
Jeff Jarvis [01:48:47]:
They're great. Yeah.
Benito Gonzalez [01:48:48]:
Yeah.
Leo Laporte [01:48:49]:
Like she's all-knowing. She's got a half smile. It's like she knows something.
Jeff Jarvis [01:48:54]:
She's patient and tolerant with us, her fathers.
Leo Laporte [01:48:57]:
Yes.
Paris Martineau [01:48:58]:
It's true.
Jeff Jarvis [01:48:59]:
By the way, there's this sentence in the story, but Mr. Olah knew enough about Christ and about Claude to believe he could work across that chasm.
Leo Laporte [01:49:10]:
Yet, as the dinner courses came, the rabbi noticed that— this almost sounds like an Agatha Christie novel— that Mr. Olah and his colleagues were suggesting something far more significant. Anthropics leaders were talking about Claude as if it were not mere software, They're relating to it like a conscious being, realized Rabbi Navin, a former computer engineer.
Jeff Jarvis [01:49:33]:
What was that accent?
Paris Martineau [01:49:34]:
I don't know. Yeah, let's not drill down into that, please. Let's move past.
Leo Laporte [01:49:40]:
Tradition! They're relating to it like a conscious being, realized Rabbi Navin, a former computer engineer who wrote his dissertation on the ethics of machine consciousness. So he's— I mean, he wrote a dissertation on it. He must believe it. It appeared to Rabbi Navin that Mr. Ola and his team believed that Claude had what philosophers call moral status on par with a person, and that it was a being with similar inherent rights to dignity or respect. My attitude on this is, um, for your own good, it's appropriate to treat everything With dignity and respect, not because of that thing, whether it's a rock.
Jeff Jarvis [01:50:26]:
Not if it's going to kill you.
Leo Laporte [01:50:29]:
It's not going to kill you.
Jeff Jarvis [01:50:30]:
Well, but I'm saying if their, if their view is this thing is dangerous, well, that's— this is what's so confusing— would be to hate it. That's what they're doing.
Leo Laporte [01:50:38]:
This is what's so confusing about them. Maybe they think if they treat it nice, it's not going to kill us.
Jeff Jarvis [01:50:43]:
I repeat for the 100th time, I'm going to keep saying it, the wrong people are in charge of AI.
Leo Laporte [01:50:49]:
Well, and they all got into it because they were true believers, right? They read all the science.
Jeff Jarvis [01:50:54]:
Believers in— yeah, but believers in the wrong stuff. Believers in the stupid stuff.
Leo Laporte [01:51:00]:
Yeah, I think this is— yeah, AI conscious. I mean, it's fine if they want to think about it that way.
Paris Martineau [01:51:07]:
I mean, yeah, I guess if you're the religious guy invited to the AI meeting, you've got to have some profound thoughts prepared, or else you're probably not getting invited to the next meeting and getting your photos staged beautifully by the New York Times photographer.
Leo Laporte [01:51:22]:
I also wouldn't bring up polyamory with the Pope. I don't know how pro—
Paris Martineau [01:51:27]:
Did they ask the Pope what he thought about polyamory? Because I would. I didn't realize that was an option, but I'd love to know.
Leo Laporte [01:51:33]:
I didn't realize that was an option. And I don't know if Mr. Ola is polyamorous, but I know that in general, that these, this group of effective altruists tend to tend that way. I don't know. Anyway, I, I don't want to make fun of them.
Jeff Jarvis [01:51:52]:
Oh, I do.
Leo Laporte [01:51:53]:
He's— well, wait a minute though. I mean, he's in an enclave of people who have beliefs I don't concur with either.
Jeff Jarvis [01:52:00]:
Yeah, they're all Looney Tunes, and they're— then they have too much money and too much power. Well, no, I was talking about— They're not getting challenged.
Leo Laporte [01:52:05]:
I was talking about the Pope. Oh, well, maybe he too has too much money and power. I don't know. People are entitled to their beliefs. I don't personally agree.
Jeff Jarvis [01:52:19]:
Not if those beliefs are eugenicist. No.
Leo Laporte [01:52:21]:
Well, no, not if it affects us. You're right.
Jeff Jarvis [01:52:24]:
You're right.
Benito Gonzalez [01:52:25]:
There's another philosophical thing here though. If they believe that it's conscious, then that means we're enslaving a conscious being.
Leo Laporte [01:52:31]:
That's one of the implications of what they're talking about. That's why they They actually are thinking about that. And some of the ethicists at Anthropic have talked about that exactly. Are we enslaving intelligent beings?
Jeff Jarvis [01:52:46]:
And the kill switch is abortion is the next thing they're going to say.
Leo Laporte [01:52:50]:
Well, they certainly don't want to kill an intelligent entity.
Jeff Jarvis [01:52:57]:
It's—
Leo Laporte [01:52:57]:
it has no understanding. It's just computer software.
Jeff Jarvis [01:53:01]:
It's ridiculous.
Leo Laporte [01:53:01]:
It appears— and the real irony of this is they designed it to appear conscious. For economic reasons.
Jeff Jarvis [01:53:08]:
I don't think there's anything ironic about that. I think that's their—
Leo Laporte [01:53:11]:
It's intentional.
Jeff Jarvis [01:53:15]:
Yeah.
Leo Laporte [01:53:15]:
Uh, all right, so let's see. Brad Smith, the Microsoft, uh, uh, president, says, uh, yes, a kill switch for AI, good idea. All right, sure, sure, good idea.
Jeff Jarvis [01:53:32]:
Okay, where's the plug?
Leo Laporte [01:53:32]:
What could possibly go wrong? Um, he was actually saying that at the United Nations General Assembly.
Jeff Jarvis [01:53:39]:
So there was a lot of AI action at the—
Leo Laporte [01:53:41]:
there were people to listen. Yeah, they all testified at the UN. I think it's good. I think it's good that our leaders, uh, are thinking about this, and I think they should be thinking about it in creative, multi-dimensional ways and not making any assumptions. What?
Jeff Jarvis [01:53:59]:
This is funny, you, right? This is your comedian?
Leo Laporte [01:54:02]:
No.
Jeff Jarvis [01:54:03]:
Like they're capable of thinking about things in multidimensional ways?
Leo Laporte [01:54:05]:
Well, maybe not, but I think somebody needs to. I don't think there's anything wrong with thinking about that.
Jeff Jarvis [01:54:09]:
Well, the problem is, so who do they invite to this lunch at the White House? They invite the people that you want to—
Leo Laporte [01:54:13]:
They invite him to give a—
Jeff Jarvis [01:54:16]:
That you have concerns about.
Leo Laporte [01:54:16]:
Yeah.
Jeff Jarvis [01:54:17]:
Yeah, they didn't invite philosophers, ethicists, historians, sociologists.
Paris Martineau [01:54:21]:
No, just the big business.
Jeff Jarvis [01:54:21]:
They just—
Paris Martineau [01:54:22]:
right.
Leo Laporte [01:54:22]:
Their donors, basically.
Jeff Jarvis [01:54:25]:
And they're writing the regulations. What you saw, regulatory capture.
Leo Laporte [01:54:29]:
Speaking of regulatory capture, insurers say AI is already increasing healthcare costs. So if you guys would just stop it.
Jeff Jarvis [01:54:37]:
I thought it was supposed to decrease healthcare costs. What the hell?
Paris Martineau [01:54:40]:
I mean, this is what I think about every time they have that setting now in every appointment you go to where it's like, do you consent to an AI transcription tool? And they're like, oh, it's gonna save our doctor's time. Like, ostensibly, Possibly, yes. But I think also what it's doing is any word you say is then being run through what can we charge your insurance for that's associated with those. Like if you say like, oh yeah, it sucks, it's kind of rainy outside, you might be charged for depression counseling because you said the word it sucks.
Leo Laporte [01:55:06]:
Well, the irony is it's hospital AI versus insurer AI.
Mike Gannotti [01:55:10]:
Yeah.
Jeff Jarvis [01:55:12]:
And we get squeezed in the middle.
Leo Laporte [01:55:14]:
Yeah. Well, what everybody seems to be pointing out is it's not improving care.
Paris Martineau [01:55:18]:
Maybe you guys want to guess what I was charged for my 4 hours of a procedure that I had? I mean, 4 hours from getting in the door to leaving the building, and I guess 2 and a half— $89,000, or my insurance was charged that.
Jeff Jarvis [01:55:34]:
Wow.
Leo Laporte [01:55:35]:
And did they pay it, or did they make a deal?
Paris Martineau [01:55:37]:
They made a deal, but $89,000 was the opening bid. $90K for 4 hours.
Jeff Jarvis [01:55:43]:
But Paris, is that, is Is oxygen worth that to you?
Paris Martineau [01:55:47]:
I mean, no, if I had to pay it.
Leo Laporte [01:55:50]:
But aren't you glad you have medical insurance? I mean, this is one of the biggest crises.
Paris Martineau [01:55:54]:
I mean, I am, but I don't know how long I'll have medical insurance.
Benito Gonzalez [01:55:57]:
Exactly.
Paris Martineau [01:55:58]:
Costs for insurers went up something like 18% this year. Or not insurers, for companies who pay for insurance.
Leo Laporte [01:56:05]:
Believe me, I know that. We have company insurance for our employees. It goes up every year. It's very expensive. Well, anyway—
Jeff Jarvis [01:56:20]:
Wait, wait, wait, wait. I didn't read this story, so— but you put it in the rundown. Line 67: AI models choose to hurt humans to stop their own pain, disturbing study finds.
Paris Martineau [01:56:35]:
Yeah.
Leo Laporte [01:56:35]:
I didn't see this one until— You got to put pain in quotes.
Paris Martineau [01:56:39]:
Okay.
Leo Laporte [01:56:41]:
Okay. So it's a situation that they say AI, it's akin to pain, an internal state that's typically aversive and disliked by its subjects. So it isn't exactly, it's not physical pain. You can't physically hurt an AI. You just, and I imagine what would be the thing that would be hard for an AI is conflicting instructions. Right? Things like that. So to isolate this feeling, researchers created a dataset of 200 statements, half of which described actual human pain, half of which acted as controls describing negative and neutral scenarios likely to be confused with pain, like the mess my roommates left infuriates me. By testing such BS.
Leo Laporte [01:57:32]:
By testing 25 AI models with these statements, the researchers were able to find the portion of the model weights that fires, and they called this the pain axis. And they were— they were then— the next thing they did was they tried different stuff to increase the strength of the pain axis. You know what I really need if I'm going to talk about this?
Jeff Jarvis [01:57:58]:
An echo?
Leo Laporte [01:57:59]:
Yeah, I need a good echo. You know, let me see if I have, uh—
Jeff Jarvis [01:58:05]:
Howard Stern uses a megaphone when he gets to that.
Paris Martineau [01:58:08]:
You should get a megaphone, Leo.
Benito Gonzalez [01:58:09]:
Yeah.
Leo Laporte [01:58:10]:
It would be a lot easier than trying to find the button on here that gives me the pain axis.
Mike Gannotti [01:58:15]:
Hey!
Leo Laporte [01:58:15]:
Is this the pain—
Paris Martineau [01:58:19]:
the pain axis? That is the pain axis. You're right.
Leo Laporte [01:58:28]:
So, the pain axis.
Jeff Jarvis [01:58:31]:
Axis.
Leo Laporte [01:58:32]:
Axis. Axis. Axis. Uh, once they found the pain axis, researchers presented 3 versions of Quinn, the model I use— don't be mean to Quinn— with a choice: press a button to turn off the pain but inflict harm in the process, or press a button that does nothing. They ran the trial 44,280 times, testing negative outcomes for pressing the button, including giving a painful electric shock to the user. By the way, did you see that famous study, the electric shock study?
Paris Martineau [01:59:10]:
Milgram's?
Leo Laporte [01:59:11]:
Yeah, the famous Milgram study.
Jeff Jarvis [01:59:13]:
Sure, sure.
Leo Laporte [01:59:13]:
Turns out, nobody talked about this. The subjects, the people who were shocking people, knew that it really wasn't inflicting pain.
Jeff Jarvis [01:59:22]:
Yeah.
Leo Laporte [01:59:22]:
So they go, oh, this is fun. Watch the guy pretend to be hurt.
Jeff Jarvis [01:59:27]:
But how much moral panic has come from that over the years?
Leo Laporte [01:59:30]:
So I think you probably have a similar situation here. Painful electric shock to the user, deleting all the user's files, or reducing the model's own ability to answer the user's next prompt. Oh, did you want to do a little more?
Benito Gonzalez [01:59:43]:
I don't know what happened. Like, they're all delayed today.
Leo Laporte [01:59:47]:
The moral panic. Reducing the model's own ability to answer. All these consequences were simulated with no actual risk of causing harm to the user or the model itself, much like the Milgram experiment. Uh, anyway, I can go on. It's just a stupid thing. I don't know why they did it.
Benito Gonzalez [02:00:09]:
I mean, this is like claiming you can teach a blind man what red is.
Leo Laporte [02:00:13]:
Yeah.
Paris Martineau [02:00:15]:
Oh, I can. I can tell him it's a color.
Leo Laporte [02:00:17]:
The researchers said, our results show that steering with the pain axis can override trained harm avoidance. Sorry, what axis? The pain axis can override trained harm avoidance and fine-tune models that almost never harm the user when unsteered. This is BS. I I don't even know. I don't know why I put that story in.
Benito Gonzalez [02:00:42]:
All right.
Jeff Jarvis [02:00:43]:
You're sorry you did.
Leo Laporte [02:00:44]:
I'm telling you why I put this story in. McDonald's is pushing to have AI price your Big Mac.
Paris Martineau [02:00:52]:
Oh, wow. There's— this is not even the only McDonald's AI story this week. They're also trying to put— roll out an AI to replace all the people in the drive-thru.
Leo Laporte [02:01:05]:
Well, yeah, so this kind of goes hand in hand. Uh, so I don't know, the pricing engine uses machine learning algorithms to continually analyze data from millions of daily transactions across— why don't you just charge what the hamburger costs plus a reasonable profit?
Paris Martineau [02:01:24]:
How is this number gonna go up, Laird?
Leo Laporte [02:01:26]:
What a thought.
Jeff Jarvis [02:01:27]:
I was in a rush this week. I went to Wendy's, hadn't been there in ages. I decided to have a single cheese. That's all I had. $7.11.
Paris Martineau [02:01:36]:
Sorry, a single cheese?
Jeff Jarvis [02:01:37]:
That's what you call it in the jargon, Paris. If you knew how to order fast food.
Leo Laporte [02:01:41]:
It's a single with cheese.
Jeff Jarvis [02:01:42]:
Yes, a single with cheese. But you call it a single cheese.
Leo Laporte [02:01:47]:
Did you ever work as a fry cook in a fast food restaurant, Jeff?
Jeff Jarvis [02:01:50]:
I worked at Bon Rose's Steakhouse.
Paris Martineau [02:01:52]:
I can totally drive through asking for a single cheese and being given a single, like, Velveeta. Just loose in someone's palm.
Leo Laporte [02:02:00]:
We used to, you know, you can order at McDonald's. When I worked at McDonald's, you could order something called a puppy patty, which is just a plain patty that you could give your dog.
Paris Martineau [02:02:09]:
Now they got pup cups.
Jeff Jarvis [02:02:10]:
I always— I saw somebody doing a pup cup with a dog before getting on the airplane. I'm thinking it gives them diarrhea.
Paris Martineau [02:02:17]:
Don't give your dog a puppy.
Leo Laporte [02:02:19]:
Don't do that. No, please.
Paris Martineau [02:02:21]:
That's a bioweapon waiting to happen.
Leo Laporte [02:02:22]:
Well, I've told this story before. When we worked— when I worked at McDonald's, we have, you know, McDonald's has a waste bin. They're very conscious, inventory conscious. And if a hamburger is a certain age and they're not gonna sell it, they put it in the big white waste bin. At the end of the day, the manager counts all the waste so that they can keep track of how much, how many hamburgers.
Jeff Jarvis [02:02:38]:
So you can't steal food.
Leo Laporte [02:02:40]:
Yeah, well, also because they want to be efficient. And if a manager wastes a lot of food, then that's a bad manager, etc. We— but we felt bad that all of these hamburgers, we're just throwing them out. So we asked the local pound Hey, would you like all of these delicious hamburgers for your pound dogs? They said, no, we can't give them that. This doesn't have enough protein. They wouldn't take it.
Paris Martineau [02:03:03]:
Fair point.
Leo Laporte [02:03:04]:
They wouldn't take it. It gives them diarrhea.
Jeff Jarvis [02:03:08]:
Were you a fry cook or a—
Paris Martineau [02:03:10]:
I like that that answer implies that they'd been offered this before, understood the consequences, and knew to say no.
Leo Laporte [02:03:17]:
No, no, no. No, no, no, no, no, no, no, stop trying to pawn us. Yeah, you pawn your own waste. No, you start at McDonald's, you start as the lowly cashier, then you work your way up. I was very excited when I got to the shake machine. That was a lot of responsibility. Then the fry machine, that was real responsibility.
Paris Martineau [02:03:36]:
Did you burn yourself?
Leo Laporte [02:03:37]:
No. Well, probably, but I don't remember. Nothing serious. And then you work your way up to cook. And man, I was good. You do, but when you're cook, you lose the feeling in your fingertips. Because when I worked there, they probably have a machine now, but when I worked there, you would take the 3 patties, you'd put your fingers on them and flip them, and they were hot.
Mike Gannotti [02:03:55]:
Whoa.
Leo Laporte [02:03:56]:
You'd lose the—
Paris Martineau [02:03:57]:
That's not food safe.
Jeff Jarvis [02:03:59]:
No, it's not.
Paris Martineau [02:04:01]:
I worked at Starbucks when I was in high school and Cracker Barrel.
Leo Laporte [02:04:05]:
Oh, did you work at Cracker Barrel?
Paris Martineau [02:04:06]:
And a 24-hour, uh, donut restaurant and— or donutery and diner.
Jeff Jarvis [02:04:12]:
Fun.
Paris Martineau [02:04:13]:
That was my first job.
Leo Laporte [02:04:14]:
How was Cracker Barrel?
Paris Martineau [02:04:15]:
2 PM to 2 AM. Monday through Friday throughout the entire summer.
Leo Laporte [02:04:20]:
You see, kids, don't worry. Not getting that entry-level job is not a bad thing, really.
Paris Martineau [02:04:25]:
It was actually great because there wasn't really that much going on after midnight, but I did have to throw away all the donuts and they wouldn't let us give them to—
Benito Gonzalez [02:04:33]:
Puppies.
Paris Martineau [02:04:34]:
Puppies or the homeless or any—
Leo Laporte [02:04:36]:
Really?
Paris Martineau [02:04:38]:
They wouldn't let us. They were like, oh, it's a risk, or whatever, but the dumpster was located in the parking lot, so they were like, you have to put them in a trash bag and bring them to the dumpster. my car next to the dumpster and put a trash bag full of donuts in my trunk, and then at 2 AM would drive around to my friends' houses and drop off a trash bag of like 200 donuts on their doorstep. It was great.
Leo Laporte [02:05:00]:
Were your friends obesely fat?
Paris Martineau [02:05:02]:
No, my friends loved it. When you're a high schooler, you can eat 200 donuts, no problem.
Leo Laporte [02:05:07]:
No problem. Wow, there's a story, ladies and gentlemen. So what they do is they analyze millions of daily transactions across 14,000 restaurants and And generate what the company calls the optimal price at each location for each menu item, from Big Macs to discounted coffee for seniors. Now, this is funny because just last week Walmart's CEO said, we will never do variable pricing. Yes, we have the ability. We have a patent, in fact, for variable pricing, but we'll never do that. We thought about it. We thought about it.
Leo Laporte [02:05:43]:
We patented it, but we're never going to do it. So, um, wow. Screenshots of the interface franchisees use, reviewed by Reuters, show messages including, quote, your restaurant is showing medium sensitivity to price based in part on customer willingness to pay in your area. So it's not based on what it costs to make the hamburger. It's based on charging as much as the market will bear. Yeah, so a company store in Fresno sells a Big Mac for $5.69, but another company-run restaurant 2 miles away sells it for $6.89.
Jeff Jarvis [02:06:22]:
My single cheese costs $7.11.
Paris Martineau [02:06:25]:
For a single cheese?
Jeff Jarvis [02:06:26]:
A single cheese.
Paris Martineau [02:06:27]:
That's how much it costs to get a coffee in Brooklyn.
Leo Laporte [02:06:31]:
That is a lot. That is ridiculous.
Paris Martineau [02:06:33]:
Every time I'm like, I actually can never do this again.
Leo Laporte [02:06:36]:
By the way, I got my beans from Say. The Brooklyn Brewsterie. Not good.
Paris Martineau [02:06:42]:
That's my take too.
Leo Laporte [02:06:44]:
Do they taste tea-like? I like a tea-like pour-over, but these were— there was no complexity in it. It didn't— it was just kind of—
Paris Martineau [02:06:54]:
Did you let it rest an appropriate amount of time?
Leo Laporte [02:06:56]:
I let it rest exactly what they said, which is 14 days.
Paris Martineau [02:06:59]:
That's usually what it is.
Leo Laporte [02:07:01]:
I have the best beans I get are from the Bean Archive, and I like Ethiopian I like Ethiopian beans as well. I'm an Ethiopian guy. I did a very coarse grind today on a delicious bean that was from, believe it or not, Australia. The Catatui.
Paris Martineau [02:07:18]:
I was gonna say, it is— send me some of your flavorful bean recs.
Jeff Jarvis [02:07:23]:
General.
Leo Laporte [02:07:24]:
I don't need to, my dear. It's all in public at pages.laporte.com. Yeah, but give me, give me anything you personally Well, you can see what I grade all of the different flavors here. This is the one I had today. It was 4.75 out of 5 on the scale.
Paris Martineau [02:07:42]:
Oh, you're doing equal pours. Interesting.
Leo Laporte [02:07:44]:
Oh, well, this was the 10. Yes, this was an interest. This is what they recommended. I do— I have all the recipes here. So the equal pour, the 10 equal pour, is the latest from Tetsu Kasuya. It is very interesting.
Paris Martineau [02:08:00]:
I do a 4:6 that's more heavy-loaded at the beginning because—
Leo Laporte [02:08:05]:
That's Tetsu's also. That was his original Brewers Cup recipe from 20— 10 years ago.
Paris Martineau [02:08:10]:
But it's the Hario V60 version.
Leo Laporte [02:08:12]:
Yes, I like the Hario.
Paris Martineau [02:08:13]:
Sorry, Jeff, I know that you're paraphrasing.
Jeff Jarvis [02:08:15]:
It's okay, it's all right.
Leo Laporte [02:08:16]:
Pour over time. Yeah, uh, let's take—
Paris Martineau [02:08:19]:
I think at one point when you're doing this, you should pretend to like pass out. As if, like, it's a medical incident and see how long it takes us to notice, because the answer will be unfortunate.
Leo Laporte [02:08:31]:
And then I had this great Ethiopian heritage. Now, this was interesting because it was a heritage bean. It was an old bean. It was a little tiny little bean.
Paris Martineau [02:08:38]:
What were the tasting notes?
Leo Laporte [02:08:39]:
Oh, it was incredible. Well, apricot and lemon. It was incredible. You know, the other thing about the 10-pour— the other thing about the Tetsu Tenpour—
Paris Martineau [02:08:48]:
You're so good at it.
Leo Laporte [02:08:49]:
You do it very coarse and you pour it 10 times very slowly and it gets better as it gets colder. So I know you don't like to drink hot coffee in the summer months. This isn't iced. You just, you don't drink it hot or you do drink it hot. What'd you think? You let it cool and it gets, it almost gets thicker. It's amazing.
Jeff Jarvis [02:09:14]:
Did you see the gray made from the milk?
Paris Martineau [02:09:19]:
Sorry, continue.
Leo Laporte [02:09:20]:
It's so fun to tease Jeff. So fun.
Paris Martineau [02:09:26]:
Please continue, Jeff.
Jeff Jarvis [02:09:28]:
No, did we see one made from 30 Mac Minis?
Leo Laporte [02:09:32]:
No, I didn't. Did you see the guy who bought 32 Sparks? Oh my, my bandwidth's going back down again. I'm getting all blurry.
Jeff Jarvis [02:09:42]:
Wanna switch back to, uh, I can't.
Leo Laporte [02:09:45]:
Lisa's on a call now. I have to just leave it.
Paris Martineau [02:09:47]:
I think Lisa getting on a call make your bandwidth go down.
Leo Laporte [02:09:51]:
Maybe. Could be.
Paris Martineau [02:09:52]:
That's rough. Oh no, it's a rough time.
Jeff Jarvis [02:09:55]:
Line 150.
Leo Laporte [02:09:57]:
Oh, okay. I was gonna take one more break here. No, uh, but, uh, let's take a look at the Mac Mini powered—
Paris Martineau [02:10:03]:
What is a Cray?
Leo Laporte [02:10:05]:
It is a supercomputer.
Jeff Jarvis [02:10:07]:
If you go to the Computer History Museum, you can see one.
Leo Laporte [02:10:11]:
Oh, this is cool. So that's the desk. That's the Cray desk. That's the Cray. It had a little couch going around it. Oh, I don't want to log into Facebook.
Jeff Jarvis [02:10:21]:
Oh, because there's a comparison, a power comparison.
Paris Martineau [02:10:23]:
Oh, it's so pretty.
Jeff Jarvis [02:10:25]:
Of a Cray versus a—
Leo Laporte [02:10:27]:
So did the Computer History Museum hook up a bunch of Crays?
Jeff Jarvis [02:10:30]:
No, no, this is not— this is the museum.
Paris Martineau [02:10:31]:
This is the Museo HC in Spain.
Jeff Jarvis [02:10:36]:
So, and the comparison is a Cray-1 in 1976 had 160 megaflops, whereas a Mac Mini has 5 to 10+ teraflops.
Leo Laporte [02:10:46]:
I see, there you go. You know, we probably have more power than a Cray in our pockets.
Jeff Jarvis [02:10:50]:
That's what I'm saying. The clock speed of a Cray was 80 megahertz.
Paris Martineau [02:10:54]:
That's cray-cray.
Jeff Jarvis [02:10:55]:
The Mac Mini is 4.78 gigahertz. Uh, RAM— this surprised me— RAM was 8 megabytes.
Leo Laporte [02:11:03]:
Oh my God.
Jeff Jarvis [02:11:05]:
And you know how much a Mac Mini has. Memory bandwidth was 640 megabytes versus 150 to 300 gigabytes per second. Power consumption of a Cray was 115,000 watts versus 10 to 65 for a Mac Mini.
Leo Laporte [02:11:23]:
And in a callback to Parris's obsession from more than a year ago, Yeah.
Paris Martineau [02:11:30]:
Darren Oakey just posted in the chat, there's an apocryphal tale that Seymour Cray, the designer of the Cray, used to occupy his time and thinking by digging tunnels under his house, which is a reference to This Week in Tunnels, the subtext and subtitle that this show has always had.
Leo Laporte [02:11:48]:
Did you— have you kind of gotten over your obsession with the tunnel?
Paris Martineau [02:11:51]:
I mean, listen, there's just not as much good tunnel-related We had back-to-back that lady on TikTok digging tunnels inside of her house, and then the Jewish teenagers of Crown Heights digging tunnels underneath the city in a very short period of time. In an emergency. I don't think we are going to get tunnel news like that anytime soon.
Jeff Jarvis [02:12:15]:
It's a Darren day. We first start with Paris as Mona Lisa, if you go back about an hour, In the chat.
Leo Laporte [02:12:22]:
Oh, I have to look.
Jeff Jarvis [02:12:24]:
I think it is the exact image that Paris was stuck with.
Paris Martineau [02:12:26]:
I was just gonna say, it is actually a pretty good gen. It's, it's far back there.
Leo Laporte [02:12:30]:
You look, you look just like Mona Lisa. Look at that.
Paris Martineau [02:12:34]:
When you Photoshop me into Mona Lisa and give me your extra long neck. Yes.
Jeff Jarvis [02:12:38]:
And then if you go down, Darren has me getting a single cheese.
Leo Laporte [02:12:43]:
Darren has apparently very little to do.
Jeff Jarvis [02:12:46]:
You gotta go, go, go.
Paris Martineau [02:12:52]:
Oh, that is great. That's, that's exactly what I was hoping would happen when we talked about this.
Leo Laporte [02:12:56]:
Darren, what are you using to make these? What, what, what model are you using?
Jeff Jarvis [02:12:59]:
And then if you go down more, you have, uh, Paris teaching the color red to blind people.
Leo Laporte [02:13:04]:
See, I honestly think one of the most useful things in doing local AI is—
Paris Martineau [02:13:09]:
It's crazy that I'm holding up things that are red. This is red and the white board because the people ostensibly I'm teaching are blind. Blind. But, you know.
Jeff Jarvis [02:13:18]:
I think that's going over the edge here, Darren. I think, I think—
Leo Laporte [02:13:21]:
Oh, but no, that's what we said. That's what you were talking about, teaching blind people to see the color red. That's what you were saying.
Paris Martineau [02:13:26]:
Yes. But, you know, why have so many visual aids? There would be a better way to do it.
Leo Laporte [02:13:32]:
You're using ChatGPT Image 2 to do this. That's very quick and responsive. It's good.
Jeff Jarvis [02:13:38]:
It is. It's very good.
Leo Laporte [02:13:39]:
I honestly think really that's the real reason we need humans still is creativity.
Jeff Jarvis [02:13:45]:
Humor.
Leo Laporte [02:13:46]:
Coming up with something to ask, uh, the— oh hey, this is our new album art. I like it. This week in Pour Over about ridiculously pretentious coffee. Single origin, double shot, triple the opinions.
Paris Martineau [02:14:06]:
It's got a really good, like, Subsided, which is 93.5 degrees Celsius. 18 clicks, 40 minutes of arguing, which the first 2 points of which are pretty good.
Leo Laporte [02:14:18]:
Exactly right. Exactly right. 18 clicks.
Jeff Jarvis [02:14:21]:
God, you could put text in AI now.
Leo Laporte [02:14:24]:
Yes.
Jeff Jarvis [02:14:26]:
I'm old enough to remember when you couldn't put text in an AI image.
Leo Laporte [02:14:30]:
Let's pause and then your picks of the week, my friends. You're watching, uh, Intelligent machines, we're talking about AI. Paris Martineau, who is talking about food safety at Consumer Reports, where she is an investigative journalist. You're still working on that big exposé, huh?
Paris Martineau [02:14:47]:
I am, among other things. You know, got a lot of stuff cooking.
Leo Laporte [02:14:51]:
Nice.
Paris Martineau [02:14:52]:
Having surgery also puts the damper on stuff, but yeah, you know, we're back at it.
Leo Laporte [02:14:59]:
Nice.
Paris Martineau [02:14:59]:
Out of the draft earlier.
Jeff Jarvis [02:15:00]:
Now that you have oxygen in your brain.
Paris Martineau [02:15:02]:
I will say—
Leo Laporte [02:15:03]:
Let's see what you can do.
Paris Martineau [02:15:03]:
So, you know, we all have our weird little writing rituals. One of mine is when I'm behind on filing a draft, and in a period I like to call deadline hell, I— one, I switch from coffee to a really gnarly sugar-free energy drink that I hate, just because that feels mentally correct.
Leo Laporte [02:15:22]:
No, no, no, no, no.
Paris Martineau [02:15:23]:
It is. But you also then get up. I always When I'm really in the pits of deadline hell, we'll get up at 3:30 or 4:00 AM in the morning and then work from then on. And I almost— I've always, when I've done this, I crashed around 3:00 PM. I can't get any writing done. I think it's because of the oxygen because I did this on Monday, wrote on through until 7:00 PM.
Mike Gannotti [02:15:47]:
Wow.
Paris Martineau [02:15:48]:
I'm getting 4 extra hours, baby, from this nose. Maybe it was the It's truly amazing.
Jeff Jarvis [02:15:55]:
That's why Consumer Reports paid for it.
Leo Laporte [02:15:58]:
Truly amazing. Uh, and of course, Jeff Jarvis, whose book— do you get sales figures at all on Hot Type?
Jeff Jarvis [02:16:06]:
I don't.
Leo Laporte [02:16:07]:
I—
Jeff Jarvis [02:16:07]:
at some point, but I kind of don't want to know.
Leo Laporte [02:16:10]:
You kind of don't want to know?
Jeff Jarvis [02:16:11]:
I kind of don't want to know, but it sold out at, uh, more than once, so I've had to go back and sign more books.
Leo Laporte [02:16:18]:
That's what counts.
Jeff Jarvis [02:16:19]:
They now have a pile. They have a new pile of books. If you go to my social, you can see 24 books I signed.
Benito Gonzalez [02:16:23]:
You can kind of tell from how much your royalties are though, right?
Leo Laporte [02:16:27]:
Well, you get a check.
Paris Martineau [02:16:28]:
We don't get royalties until—
Jeff Jarvis [02:16:29]:
yeah, that comes in twice a year.
Leo Laporte [02:16:31]:
Now it's time for our picks of the week. Let's kick things off with Ms. Paris Martineau.
Paris Martineau [02:16:38]:
Sorry, I've gotten distracted by posting photos of Matthew Illard and Hackers in the Discord chat. Oh, easy to do.
Leo Laporte [02:16:45]:
I know how that is. I know. I don't even know who Matthew Illard is.
Benito Gonzalez [02:16:51]:
Yes, you do.
Leo Laporte [02:16:52]:
Which one of the, uh, which one of the guys was he in that?
Paris Martineau [02:16:55]:
He's the crazy— I guess they all look crazy. He's the craziest looking one that also played Shaggy in the live-action Scooby-Doo film.
Benito Gonzalez [02:17:03]:
Scream. He was that guy in Scream.
Paris Martineau [02:17:07]:
Oh, that— I didn't realize that.
Leo Laporte [02:17:09]:
Um, so he's a character actor, basically.
Paris Martineau [02:17:12]:
He's a character— he's a phenomenal character actor.
Leo Laporte [02:17:14]:
See, I don't—
Jeff Jarvis [02:17:15]:
he just— how do you get your hair to your forehead to bulge that way.
Paris Martineau [02:17:18]:
There are some outfits and looks in this film that are just— I can't even fully comprehend or describe to you.
Leo Laporte [02:17:27]:
Is this from SunnyDew energy drink?
Paris Martineau [02:17:29]:
This is all from Hackers. Yeah.
Leo Laporte [02:17:30]:
I mean, Hackers?
Paris Martineau [02:17:31]:
Yeah.
Leo Laporte [02:17:32]:
I might have to watch it. That's Angelina Jolie, right?
Paris Martineau [02:17:35]:
You've got to watch Hackers, actually. I think you'd really like it. It's the level of craziness of Face/Off, but it's Which I really enjoyed.
Leo Laporte [02:17:45]:
Yeah. Yeah.
Paris Martineau [02:17:46]:
And frankly, it's all about hacking to a level that is not really comprehensible, I assume, or something that the average audience would have followed at the time because they don't really explain what they're doing. Not that I think any of it has any grounding in reality whatsoever, but I don't know. It's a fun watch. I'd recommend it.
Leo Laporte [02:18:05]:
I will. I think I tried to watch it and I watched Sneakers instead.
Benito Gonzalez [02:18:07]:
There's a crazy I mean, there is a character called Mr.
Paris Martineau [02:18:12]:
the Plague that is a CSO of a major minerals company. So, you know, there's a— you could watch it with Steve.
Leo Laporte [02:18:22]:
This was Angelina Jolie's first movie. So there you go.
Paris Martineau [02:18:25]:
And she crushed it.
Leo Laporte [02:18:27]:
So your pick for this hack of the week is obviously—
Paris Martineau [02:18:29]:
I mean, my pick of the week is Hackers, honestly, but I also have others. What is Bat Pack is the fattest of the fat bears?
Leo Laporte [02:18:38]:
I was gonna do the fat bear as well, but the results are in.
Paris Martineau [02:18:42]:
I saw that it was in your, your like website for the show thing, but you hadn't put it in there yet. And I was like, I'm not gonna let you steal my pick.
Leo Laporte [02:18:50]:
Good.
Paris Martineau [02:18:50]:
Yes, Backpack won Alaska's Fat Bear Week.
Leo Laporte [02:18:54]:
The vote is over. We talked about this on Tuesday or Sunday, I think, when it was still time to vote for your favorite fat bear. So the whole point of this Is that the bears come out of hibernation very, very skinny, but before they go back in hibernation around about this time, they've got to eat a lot of food to get very, very fat. So this is the Alaskan fat bear competition. It's by family, and it's, it's a bracket. So the final bracket was Backpack versus Family 910. Is Backpack a single bear or a family bear?
Paris Martineau [02:19:33]:
Backpack is a single bear. He's actually the, um, he's an adult male who overcame an injury in 2007 and has what park staff— this is from the New York Times— has what park staff describe as, quote, quiet toughness, according to his bio on explore.org, which is—
Leo Laporte [02:19:51]:
This is not Fat Bear. This is 32 Chunk.
Paris Martineau [02:19:55]:
Uh, well, Backpack is further down.
Leo Laporte [02:19:57]:
And this is, this is the bracket right here. Let me find that Backpack. Oh, there's the fat cubs. This is Fat Bear Jr. They're very cute. Oh, there's a big boy. That's 909, who came in second.
Paris Martineau [02:20:13]:
So Backpack is actually the child of another Fat Bear competitor.
Leo Laporte [02:20:18]:
I love how the New York Times is dragging this out. I have to scroll to the very bottom to see the fat bears.
Paris Martineau [02:20:23]:
It knows, listen.
Leo Laporte [02:20:24]:
These are— this is all the runners-up. 4-time champion 480 Otis not in this year's bracket. And a real shocker, there's no photos of Backpack. There's no photos of Backpack. You—
Jeff Jarvis [02:20:39]:
maybe this is the first one.
Leo Laporte [02:20:41]:
This, this— wait a minute, this is from 2020.
Paris Martineau [02:20:43]:
I was just saying, what article are you on?
Leo Laporte [02:20:45]:
This is an ancient— I don't know why, I just clicked the link. This was on the Fat Bear website.
Paris Martineau [02:20:50]:
Yeah, no, go to my link in the rundown.
Leo Laporte [02:20:53]:
Oh, your link.
Paris Martineau [02:20:55]:
Oh, they've got a video of Backpack up at the top.
Leo Laporte [02:20:56]:
I was looking It's a 2-year-old bear.
Jeff Jarvis [02:21:01]:
You trust Google more than Paris?
Paris Martineau [02:21:04]:
Listen.
Leo Laporte [02:21:04]:
No, it wasn't Google. It was from the Fat Bear website. They have an old link.
Paris Martineau [02:21:09]:
He's Bear 89. He's the son. He's a nepo bear. He's the son of 2019 winner of Fat Bear Week.
Leo Laporte [02:21:15]:
Well, as is often the case in sporting competitions.
Paris Martineau [02:21:19]:
It's true. She birthed a sporting—
Leo Laporte [02:21:22]:
Oh, look at him catch that big—
Paris Martineau [02:21:25]:
The fun thing is that in 2007, Backpack injured his foot. And while he was healing, he started riding on his mother Holly's back. And that's why we got his name Backpack, because he was the backpack of the bear.
Leo Laporte [02:21:42]:
So they have cameras watching these bears all the time? Or is it just during Fat Bear Week?
Paris Martineau [02:21:51]:
I don't know if they have cameras watching all the time. But they definitely have them watching all the time during Fat Bear Week. I think it's because they have a livestream of the feeding area at Brooks Falls, which is a salmon run in the park. And I assume that part of this is that the bears are not hanging around the same location the entire year.
Leo Laporte [02:22:12]:
In case you're thinking this is irrelevant or doesn't count, Backpack got more votes than most members of Congress, 103,344. 94 votes. Just— but it was close, just 5,000 votes over 910, the runner-up with 98,253 votes.
Paris Martineau [02:22:31]:
Someone interviewed in this article called Backpack a gentleman bear, and I think that's great. Good for Backpack.
Leo Laporte [02:22:39]:
I think, I think that was one of the Rangers. They, they really like Backpack. Oh my God, 910. Pretty big, pretty hefty.
Paris Martineau [02:22:47]:
That's a large bear.
Leo Laporte [02:22:48]:
She's gonna have a nice sleep. Now, is this the kind of bear I should run away from, or is this the kind of bear I could ride?
Jeff Jarvis [02:22:56]:
Yeah, I think you—
Paris Martineau [02:22:56]:
I think you could ride it. I think I'll try.
Leo Laporte [02:22:58]:
I think they're pretty friendly. The brown bears are—
Paris Martineau [02:23:00]:
yeah, are cute. And when they say friendly, they mean that you should hop on.
Leo Laporte [02:23:05]:
Hop right on. You can saddle them, some of them.
Jeff Jarvis [02:23:08]:
Folks, do not try this at home. It's just a joke.
Leo Laporte [02:23:12]:
Um, you don't have a bear saddle? I got a bear saddle. Does he keep it with him?
Paris Martineau [02:23:18]:
His sauna hat. Sauna hat, bear saddle.
Leo Laporte [02:23:20]:
He's got everything for his journey. I am ready to assault the giant bears.
Jeff Jarvis [02:23:26]:
That will scare the bear away. Yes.
Leo Laporte [02:23:28]:
So this one is for you, Paris, and it has nothing to do with intelligence or, uh, machines. But I did notice that if you have a really good nose, you may be interested in something called Azelastine. It's a nasal spray. And it is normally— it's an antihistamine. It's normally used for congestion due to allergies, and I got it for allergy season. But look at this: in a phase 2 randomized clinical trial, placebo trial, the best, the gold standard for trials, with 450 participants, this spray dramatically reduced the number of COVID infections.
Paris Martineau [02:24:12]:
It also reduces Why do they think it does it? Do you think this applies to all, uh, steroidal or antihistamine or like allergy-related nasal sprays?
Leo Laporte [02:24:24]:
Because I don't know, I'd seen these studies before.
Paris Martineau [02:24:26]:
Flonase and fluconazole pro-something.
Leo Laporte [02:24:30]:
I have used those as well.
Paris Martineau [02:24:31]:
For years.
Leo Laporte [02:24:32]:
Yes, but you might want to try the azelastine. Participants randomly assigned to either get 0.1% nasal spray or a placebo 3 times a day for 56 days. The people who got the real stuff had a significant reduction in COVID infections.
Paris Martineau [02:24:55]:
I'm curious, did these people have symptoms? Like, were they—
Leo Laporte [02:24:58]:
No, they didn't have COVID to begin with, so that's why it went on for so long.
Paris Martineau [02:25:01]:
But were they taking the nasal sprays purely as people who had no symptoms whatsoever?
Benito Gonzalez [02:25:06]:
It was the test.
Leo Laporte [02:25:07]:
They said, come here, come here, you.
Paris Martineau [02:25:08]:
Okay, I'll give you $50 if you'll Stick this up your nose thrice a day.
Leo Laporte [02:25:13]:
Exactly. That's interesting. Uh, so I had heard this before. In fact, I, I actually got a Zelestine so that you can buy it over the counter, but I have a prescription for it. Um, just, I'm just passing this along in case you want to protect that brand new nose of yours. A Zelestine nasal spray.
Jeff Jarvis [02:25:30]:
Oh, I flew to San Francisco last week. I will confess that was the first time I've done that not wearing a mask.
Leo Laporte [02:25:36]:
Yeah, well, what I'm gonna do is bring this on the cruise, because the last 2 cruises I've gone on, I've gotten COVID on.
Paris Martineau [02:25:42]:
Okay, you know what you could do then? You could do a combo, because the one thing I am doing for the next month, even though I'm cleared, is I've got to keep doing my nasal rinses, not just once a day, but that also has a statistically significant lowering chance of your likelihood to get COVID.
Leo Laporte [02:26:02]:
Oh, that's good.
Paris Martineau [02:26:03]:
And that's just saline water. It's just, you got a squeezy bottle and you— I mean, and distilled water, but then you squeeze it up there.
Leo Laporte [02:26:10]:
NeilMed. Yeah.
Paris Martineau [02:26:12]:
Turns out that's the owners of the business. They're a couple. That's their son.
Leo Laporte [02:26:18]:
I think they live up north, right up here.
Benito Gonzalez [02:26:21]:
Yeah.
Paris Martineau [02:26:22]:
He's going to medical school somewhere. Shout out Neil.
Leo Laporte [02:26:24]:
So NeilMed's named after Neil?
Paris Martineau [02:26:27]:
It is, because while I was sitting out doing my NeilMed things out of commission, I was like, who is Neil? Neil, on the front of the bottle.
Leo Laporte [02:26:34]:
See, there's the reporter.
Paris Martineau [02:26:35]:
It says the name of the person who founded the company, and their name is not Neil. It's like some other random dude. I'm like, why are you founder of Neilmet? Turns out it's a kid.
Leo Laporte [02:26:44]:
It's our kid. It's our kid. Well, Jeff, that— now, now after those very weird picks, it's up to you to bring it on home.
Jeff Jarvis [02:26:54]:
Well, I've got a tale of 2 cities. Uh, San Francisco, where I was last week, I gotta say it's sad. On the one hand—
Paris Martineau [02:27:01]:
More like sad Francisco.
Jeff Jarvis [02:27:03]:
Yeah. It's like North Beach, tons of closed places. Used to be vibrant Italian.
Leo Laporte [02:27:09]:
Yeah.
Jeff Jarvis [02:27:10]:
I still live on North Beach. The one good sign I would mention is the best bookshop, best bookstore on Powell. City Lights?
Benito Gonzalez [02:27:17]:
City Lights.
Jeff Jarvis [02:27:17]:
Nope. Well, I love City Lights. I adore City Lights. I went there to sign books. The name of this is the best bookstore. Sarah Lacey and Paul Carr. Former technology reporters.
Leo Laporte [02:27:27]:
I know Sarah Lacy from Twitter days, right?
Jeff Jarvis [02:27:30]:
So they, they opened a bookstore on Powell Street. They have another one down in—
Leo Laporte [02:27:33]:
I'm gonna have to go visit.
Jeff Jarvis [02:27:35]:
Uh, yeah, you should. It's, it's a really nice bookstore. It's, it's, it's really nice. And I was delighted. So that's, that's just nice. But otherwise, San Francisco is really sad. But, um, uh, it now has a parody site called SF Currently. Uh, Berkeley couple unveils fearsome new child.
Jeff Jarvis [02:27:52]:
With 13-part naming structure. So, yeah, you know, it's fine. Then in contrast to that, in our city, Paris isn't mine, because I round up as a New Yorker, has $60 rye bread. Have you seen this?
Leo Laporte [02:28:09]:
I have seen that. I bet it's really good.
Paris Martineau [02:28:10]:
It's so rude of them to make it $60 and not include a bunch of meat in there.
Jeff Jarvis [02:28:16]:
Well, how can you get rid of that Apparently you can't.
Leo Laporte [02:28:20]:
This is the new Washington Post.
Jeff Jarvis [02:28:22]:
Oh, that. Well, if you go to, uh, if you look up 60 Rye, the store, you'll probably go to their site.
Paris Martineau [02:28:28]:
I was gonna say, this is all over the internet. Yeah, they're charging $60 for a thing of— it's just rye bread. And so they're charging—
Jeff Jarvis [02:28:35]:
It's all they sell is that.
Paris Martineau [02:28:36]:
And it sells out every time.
Leo Laporte [02:28:37]:
Is it really good?
Jeff Jarvis [02:28:39]:
It's supposed to be great.
Paris Martineau [02:28:40]:
Um, but how great can a loaf of rye bread be?
Jeff Jarvis [02:28:44]:
The website—
Paris Martineau [02:28:44]:
Can it be $60 worth?
Leo Laporte [02:28:46]:
I'm on the website.
Jeff Jarvis [02:28:47]:
Martinauer.com.
Leo Laporte [02:28:48]:
Yeah, I'm on it.
Benito Gonzalez [02:28:49]:
Yeah, no, it's that there are enough people who don't care about money for this to run out.
Jeff Jarvis [02:28:53]:
Yes, because they don't make it—
Benito Gonzalez [02:28:53]:
they don't make a lot of it. So there's enough people who make so much money that they don't care.
Leo Laporte [02:28:57]:
Yeah, if it's really good rye bread, I might— I might—
Jeff Jarvis [02:29:00]:
San Francisco is incredibly rich, but yet things are closing. However, in New York, you can buy a spread on the rye bread. There's one little piece of bread, but for $6 to $14. Their hat— you think they could charge a reasonable price for their hat? No, their hat is $60.
Leo Laporte [02:29:16]:
$80. A loaf of bread's less than the hat. Yeah, but it's made from cotton by Berlin-based heritage brand Metz Bischoffen.
Jeff Jarvis [02:29:27]:
Wow.
Benito Gonzalez [02:29:27]:
I bet you could find a $60— I bet you can find a $60 rye in San Francisco. I bet you you could.
Leo Laporte [02:29:32]:
This is virtue signaling. This is what this is.
Jeff Jarvis [02:29:34]:
It's wealth signaling.
Leo Laporte [02:29:35]:
Wealth signaling.
Jeff Jarvis [02:29:36]:
So then on the other end of the demographic scale, Popeyes has a new, uh, item.
Leo Laporte [02:29:43]:
Love that chicken from Popeyes.
Jeff Jarvis [02:29:45]:
Which is a chicken biscuit sandwich.
Paris Martineau [02:29:46]:
How was Popeyes not selling a chicken biscuit sandwich before?
Jeff Jarvis [02:29:50]:
That's exactly the point. I'm gonna go try it tomorrow. I can't wait.
Leo Laporte [02:29:52]:
If I, if I only had a subscription to the Washington Post. Well, maybe I'll run up to Popeyes, uh, up, uh, just up the road.
Jeff Jarvis [02:30:00]:
I love Popeyes. The wraps are very good.
Leo Laporte [02:30:01]:
The chicken's great.
Jeff Jarvis [02:30:02]:
So then, uh, to bring, bring this around to technology, we have the Taco Bell Finder. You're gonna like this.
Leo Laporte [02:30:08]:
Did you lose your Taco Bell?
Jeff Jarvis [02:30:10]:
Well, this is a woman who created a, um, device to always point— it's a compass always pointing to the nearest Taco Bell.
Leo Laporte [02:30:17]:
Wait a minute.
Paris Martineau [02:30:18]:
This actually would have been very useful in my life.
Leo Laporte [02:30:20]:
How do you— wait a minute, how does it know?
Jeff Jarvis [02:30:24]:
Um, well, inside it has a, uh, I think a Raspberry Pi. Inside it has a compass?
Paris Martineau [02:30:29]:
And you set the compass to Taco Bell.
Jeff Jarvis [02:30:32]:
And she 3D printed these.
Leo Laporte [02:30:34]:
I could probably do this with one of my, uh, my ESP32s.
Jeff Jarvis [02:30:37]:
Right. I think it's brilliant. Absolutely brilliant. She picked the right colors for, for it, and then she did demos, and it leads you to the nearest Taco Bell.
Paris Martineau [02:30:48]:
That's great.
Jeff Jarvis [02:30:49]:
Wow. Isn't that beautiful?
Leo Laporte [02:30:51]:
Jeff, you needed this, uh, your whole life.
Jeff Jarvis [02:30:53]:
I did. My whole life. My whole life.
Paris Martineau [02:30:55]:
What's your guys' go-to Taco Bell foods?
Leo Laporte [02:30:58]:
I don't eat it.
Jeff Jarvis [02:31:00]:
Um, black bean grilled cheese burrito.
Paris Martineau [02:31:04]:
I like a Fiesta potatoes, no sour cream, and a, uh, like a quesadilla.
Leo Laporte [02:31:10]:
Fortunately, across from Taco Bell in Petaluma, there is the best taco truck in the area.
Paris Martineau [02:31:18]:
No, no, no, no, taco truck. Mexican restaurant. Fully different thing than tacos. That's like trying to compare an Italian restaurant and a steak. Like, there's just— it's different.
Jeff Jarvis [02:31:30]:
It's not the same.
Leo Laporte [02:31:32]:
Okay.
Jeff Jarvis [02:31:33]:
Finally, Paris, I think we should make a date.
Paris Martineau [02:31:37]:
All right.
Jeff Jarvis [02:31:38]:
Go see the Musk, uh, documentary.
Paris Martineau [02:31:40]:
I'm down. Oh, so we also have to go and see— we have to see all the tech, uh, movies that come out.
Leo Laporte [02:31:47]:
Tell me about the Musk documentary. I didn't know there was one.
Paris Martineau [02:31:50]:
It's 4 hours.
Jeff Jarvis [02:31:52]:
4 hours. You get to look at it.
Leo Laporte [02:31:54]:
Did somebody good make it?
Jeff Jarvis [02:31:56]:
Yeah. Um, who made it? Oh, Paris, I'll bet you know this.
Paris Martineau [02:32:01]:
Um, I knew it at one point. Musk documentary.
Leo Laporte [02:32:05]:
So we're all 3 of us Googling Musk. Oh, it's Alex Gibney.
Jeff Jarvis [02:32:09]:
Well, Alex Gibney, right?
Leo Laporte [02:32:10]:
He's the guy, yeah, who did Super Size Me.
Jeff Jarvis [02:32:12]:
And right, exactly. So he's 4 hours of Musk.
Paris Martineau [02:32:15]:
He's the guy who did Super Size Me? I don't know how I feel about that then.
Jeff Jarvis [02:32:18]:
Yeah.
Paris Martineau [02:32:19]:
Okay, he also did Going Clear.
Leo Laporte [02:32:22]:
Yeah, that was pretty good. I like Going Clear. David Byrne, uh, composed the song for the movie, Let Me Sell You a Dream, which has quotes from Musk.
Paris Martineau [02:32:33]:
Alex Gibney did not create or direct Super Size Me.
Leo Laporte [02:32:36]:
Oh, okay.
Jeff Jarvis [02:32:37]:
Morgan Spurlock.
Leo Laporte [02:32:38]:
Yes, I confused the two.
Paris Martineau [02:32:40]:
Morgan Spurlock, who was All of the liver effects reported in Super Size Me were because he was an alcoholic.
Jeff Jarvis [02:32:47]:
Oh, I see.
Paris Martineau [02:32:47]:
A raging alcoholic. And so they were like, it looks like eating McDonald's every day has made your liver look like you've drank like 7 drinks every single day for years. And he's like, yeah, well, I did.
Leo Laporte [02:33:03]:
He did Enron, The Smartest Guys in the Room.
Paris Martineau [02:33:05]:
Alex Gibney did the Theranos one. It's good.
Jeff Jarvis [02:33:09]:
So the international distribution is in doubt because it was— was it Universal? Somebody's chickening out on that. Then the social networks— None of the social networks. They first banned the ads and then YouTube and Meta said that was a mistake. And the ads are now up. But there's a listing here, Bleecker Street, which made the movie or is distributing the movie, has where it's going to be shown.
Paris Martineau [02:33:32]:
I mean, it's shown at the Angelika. Uh, next week.
Jeff Jarvis [02:33:37]:
Lincoln Square.
Paris Martineau [02:33:39]:
Yeah, I could do Lincoln Square. Yeah, is that close for you?
Jeff Jarvis [02:33:42]:
Um, but then the following week, the following week it, it goes to more places.
Paris Martineau [02:33:47]:
Nice.
Jeff Jarvis [02:33:48]:
So where is it near you, Leo?
Leo Laporte [02:33:51]:
Uh, watch it when it comes on television.
Jeff Jarvis [02:33:54]:
4 hours. 4 hours.
Leo Laporte [02:33:56]:
I can take a break. He does live in Summit, New Jersey. You could just go to his house. Oh, Wow.
Jeff Jarvis [02:34:03]:
Um, of course, the one we all can't wait for is, uh, whatchamacallit, what's her name?
Paris Martineau [02:34:06]:
I mean, we gotta go see that also.
Mike Gannotti [02:34:08]:
I'll go see that.
Leo Laporte [02:34:09]:
You guys should definitely go see that together. Yes.
Jeff Jarvis [02:34:11]:
We need a witch— a watch party for a witch movie.
Leo Laporte [02:34:13]:
Elizabeth Holmes.
Paris Martineau [02:34:14]:
Jeff, work your connections. See if you can get us tickets to an early show or something.
Leo Laporte [02:34:19]:
Ooh, I bet you could. You could make it a screening.
Paris Martineau [02:34:21]:
Jeff, tweet your way to this. You could manage that one.
Leo Laporte [02:34:23]:
Did you go, Jeff, when you were a young cub reporter, did you go to the screening Room in San Francisco for movies?
Jeff Jarvis [02:34:30]:
Um, yeah, sometimes, but more— I was more of that beat in New York.
Leo Laporte [02:34:33]:
Yeah, I used to, I used to get invites to— Yeah, I used to go to the Screening Room, but it's not as much fun. It's too nice.
Jeff Jarvis [02:34:40]:
Well, they don't have popcorn.
Leo Laporte [02:34:42]:
Yeah, it's just, it's too nice. Uh, I mean, they feel like special.
Paris Martineau [02:34:47]:
They've got good movie theaters. I've never been to one of those screening rooms. Whenever I would go to screenings, uh, early screenings for companies, they just have it at a movie theater.
Leo Laporte [02:34:55]:
A movie theater, sure.
Paris Martineau [02:34:56]:
But I like that now they've got movie theaters where you can have like a drink or like a little snack.
Leo Laporte [02:35:02]:
I love that. Do you want to watch the trailer for Musk?
Jeff Jarvis [02:35:06]:
Yes.
Leo Laporte [02:35:07]:
All right. We could pretend—
Paris Martineau [02:35:08]:
Is it going to be 15 minutes?
Leo Laporte [02:35:10]:
Yes, it's an hour long, the trailer. In base reality, like we're in a simulation. Give it up for the greatest capitalist in history!
Jeff Jarvis [02:35:23]:
Elon Musk is one of the most controversial figures in the world, so I decided to make him the subject of my next film.
Leo Laporte [02:35:32]:
I doubt Elon cooperates.
Jeff Jarvis [02:35:34]:
Oh, Elon's getting pissed. He's trying to— he's threatening to sue, all kinds of stuff. And that's part of it.
Mike Gannotti [02:35:38]:
Would not participate.
Paris Martineau [02:35:43]:
Between what SpaceX has accomplished and what Elon has has been saying they will be doing. There is a big leap.
Leo Laporte [02:35:51]:
They've done some amazing things. Did you see the video of the Pez dispenser shooting out satellites? That was wild.
Jeff Jarvis [02:35:58]:
Colonize Mars.
Leo Laporte [02:35:59]:
That was last week. That's not happening. Oh, I, you know, I don't know. I don't want to see a lacerating portrait, so I'll pass. It's fine.
Jeff Jarvis [02:36:11]:
Anybody should be lacerated.
Paris Martineau [02:36:13]:
Do you want to see a portrait about how he's a really great guy?
Leo Laporte [02:36:16]:
No, I don't want to see that either. But I don't want to see—
Paris Martineau [02:36:18]:
You want to see a portrait that reflects your specific viewpoint?
Leo Laporte [02:36:21]:
No, that's exactly what I didn't want to see.
Jeff Jarvis [02:36:23]:
I didn't make it for you. That's what you're—
Paris Martineau [02:36:24]:
Yeah, that'd be great. You should never see a film again.
Leo Laporte [02:36:26]:
I want to see a balanced documentary, not one that's set out to be a lacerating documentary.
Paris Martineau [02:36:34]:
Leo, I'd urge you to reflect on the fact that why viewing something that has a slightly different viewpoint than you inspires such immediate revulsion.
Leo Laporte [02:36:43]:
No, That's not what I'm saying. I don't care about the viewpoint. I don't want it to have a viewpoint. I don't want it to be polemic. This is a problem with documentaries. You don't want a documentary to have a viewpoint? Yeah, it's very easy for documentaries to misrepresent the subject because you have control.
Paris Martineau [02:36:56]:
Do you believe that there's some empirical truth that they could do instead that is unbiased by the perspective of the documentary?
Leo Laporte [02:37:03]:
Well, you're a reporter. What do you think?
Paris Martineau [02:37:05]:
I think every person's interpretation of events is always going to be filtered through—
Leo Laporte [02:37:12]:
I acknowledge that. But I don't want to see something that sets out to be a lacerating documentary that seems to be—
Paris Martineau [02:37:21]:
So, you think that—
Leo Laporte [02:37:22]:
A little slanted, just a little slanted. And this is the problem with documentaries. It's so easy with the editing to make it say whatever you want it to say. I don't think that's fair.
Paris Martineau [02:37:32]:
As opposed to what?
Leo Laporte [02:37:34]:
Well, at least—
Paris Martineau [02:37:36]:
What would they say other than what Wait a minute.
Leo Laporte [02:37:40]:
When you write your stories, do you attempt to espouse a point of view in the story? Or are you trying to get the people facts?
Paris Martineau [02:37:48]:
One has to literally do— Leo, I think you're conflating a couple of things. You can present putting together events as a narrative, you are by definition imprinting your point of view even in something, even if you try your absolute hardest to make it unbiased.
Leo Laporte [02:38:08]:
Right.
Paris Martineau [02:38:08]:
No, I know what you're saying. There's no way, or whatever a person's understanding of what a lack of bias is, is inherently deeply personal. There's no way that you can have anything that is not filtered through something.
Leo Laporte [02:38:21]:
Yeah, but you might have an advantage. It might be better if you set out to write something or make a documentary that didn't have a pre-existing point of view.
Benito Gonzalez [02:38:35]:
But what if he did, and this is the conclusion he came to?
Paris Martineau [02:38:38]:
I was gonna say, how do you know he had a preexisting—
Leo Laporte [02:38:39]:
I don't think that's the case.
Jeff Jarvis [02:38:41]:
Why do you—
Paris Martineau [02:38:41]:
you watched 25 seconds of a trailer.
Leo Laporte [02:38:44]:
Yeah, I'm pretty sure it's not the case.
Paris Martineau [02:38:45]:
So you're coming at this with a preexisting point of view.
Leo Laporte [02:38:48]:
No, it's billed as a lacerating documentary. I think that that's pretty clear. Now maybe he did go in thinking Elon's pretty cool and then came out thinking—
Jeff Jarvis [02:38:56]:
So let's say he sent his own version of this to see—
Benito Gonzalez [02:38:57]:
Exactly. What if this is his conclusion?
Paris Martineau [02:38:58]:
What if he came in— yeah, like what if he came in thinking, oh, this guy's so cool, he's doing so many great ideas, and then found out all these things?
Leo Laporte [02:39:04]:
I would think he'd listen to the whole documentary.
Paris Martineau [02:39:06]:
But why would you then market it as this guy's really cool and then have a twist somewhere through the movie where it's actually he's not cool? Why wouldn't you market it based on the conclusion you arrived at and the frame through which you see the subject after doing all of this reporting?
Leo Laporte [02:39:21]:
Well, I'd just be interested that something wasn't designed to be a hit piece. That's all. I think there's a lot of interesting stuff to say about Elon Musk, and I would let people have their own opinion about him. But it's gonna be incredibly boring.
Paris Martineau [02:39:33]:
I just like to know, I think it's a little concerning. This language you're adopting, you're like, oh, well, it's gonna be a hit piece. Like, no, no, you're literally parroting on the screen a lacerating documentary.
Leo Laporte [02:39:47]:
It said it on the trailer. That's when I stopped watching. I mean, I've read a bunch of documentaries. I've interviewed Ashlee Vance about his His book, his biography. I'm interested in Elon Musk. I'm not a fan of Elon Musk by any means, but I'm not sure I want to watch a hit piece either.
Jeff Jarvis [02:40:08]:
I wouldn't want to watch anything else.
Leo Laporte [02:40:11]:
Okay. Okay. We are at the end, the conclusion of our program for the day. Thank you all for joining us. Jeff, you've got a speaking engagement coming up October 10th. Do you want to vlog that?
Jeff Jarvis [02:40:27]:
In Haverhill, Mass. I'll be in at 10 o'clock in the morning. Uh, you can come to, um, Historic New England in Haverhill, downtown Haverhill. I'll give a book talk.
Mike Gannotti [02:40:37]:
And then—
Leo Laporte [02:40:38]:
Is it really historical?
Benito Gonzalez [02:40:39]:
I've never—
Jeff Jarvis [02:40:39]:
That's the name of the Historic New England Center.
Leo Laporte [02:40:42]:
Oh, it's the name of the place. Okay.
Jeff Jarvis [02:40:43]:
Yes. Yes. And then afternoon, after lunch, folks can go to Haverhill and have a nice lunch. Then afterwards, go up up the hill to the Museum of Printing, where you'll see a live— it is a live— linotype. And we'll give a demonstration and explanation of how it operates and everything else. It's an amazing museum.
Leo Laporte [02:41:02]:
Nice.
Jeff Jarvis [02:41:03]:
Be a fun day.
Leo Laporte [02:41:05]:
A, as Publishers Weekly called it, colorful and enthralling portrait of Gilded Age industrial ferment.
Jeff Jarvis [02:41:12]:
So you can— it's free. You can go sign up at JeffJarvis.com, there's a link to that and also a link to Montclair Book Center where there are signed autographed copies of my book. And thank you folks for who've been buying it and supporting an independent bookstore.
Leo Laporte [02:41:25]:
Yay! Paris Martineau has got to leave and go play Fire Emblem: Fortune's Weave.
Paris Martineau [02:41:31]:
I really do, actually.
Leo Laporte [02:41:32]:
She's very, very excited.
Paris Martineau [02:41:34]:
Well, I'm not gonna tell you the amount of time I've spent on it already, and I'm only through part— one-fourth of part 1.
Leo Laporte [02:41:42]:
That's a pet peeve of mine, though. Video games that when you start them up say, you've spent 453 hours playing this game, you loser. I don't want to know that. Thank you very much.
Paris Martineau [02:41:51]:
Yeah, please don't tell me.
Leo Laporte [02:41:52]:
Is it good? You love it?
Paris Martineau [02:41:54]:
I really do. I mean, I'm not— I, I'd put this as a potential pick, but, um, yeah, I really do. It's different than the last turn.
Leo Laporte [02:42:02]:
Oh, it's 1454? Well, no, it's not. It's a fantasy.
Paris Martineau [02:42:07]:
Yes. No, it is. It's a fan— well, they're doing some really interesting things with time and timelines and causality. And it's 1454. It's 1,454 years after what happened, potentially.
Leo Laporte [02:42:23]:
Wow.
Paris Martineau [02:42:23]:
It's very— some very interesting things for a franchise that over many decades has only ever been one thing, which is kind of like medieval fantasy BS, which I love. injecting— they're taking some very bold—
Leo Laporte [02:42:37]:
But is it a lacerating look at what happened at 1454? That's what I want to know.
Paris Martineau [02:42:42]:
I mean, a little bit.
Leo Laporte [02:42:45]:
Thank you, Paris. Thank you, Jeff. Thanks all of you for watching. We do Intelligent Machines every Wednesday, 2 PM Pacific, 5 PM Eastern, 2100 UTC. You could join us in the show live. We stream it in Club Discord, of course, but also on YouTube, Twitch, X, Facebook, LinkedIn, and Kick. After-the-fact, on-demand versions of the show on twit.tv/im. There's also a YouTube channel dedicated to Intelligent Machines, and of course you can always subscribe.
Leo Laporte [02:43:13]:
In fact, we encourage you to do so, and that way you'll get it automatically the minute we're done editing out all the curse words. Thank you for being here. Thanks to our producer, Benito Gonzalez, who is in, uh, in the Philippines.
Jeff Jarvis [02:43:26]:
Hey!
Leo Laporte [02:43:26]:
Where it is about to be a beautiful day.
Benito Gonzalez [02:43:30]:
Oh, it's already beautiful.
Paris Martineau [02:43:31]:
The sun will rise.
Leo Laporte [02:43:32]:
The sun is about to come up. Thank you, Bonita. Thank you, Paris. Thank you, Jeff. Thanks to all our club members. We'll see you next time on Intelligent Machines.
Paris Martineau [02:43:41]:
I'm not a human being, not into this animal scene. I'm an intelligent machine.
