Intelligent Machines 844: Poob Has It For You

Summary

Intelligent Machines 844: “Poob Has It For You”

TWiT.tv • November 6, 2025
Host: Leo Laporte
Co-hosts: Paris Martineau, Jeff Jarvis
Special Guest: Jeremy Berman (Post Training Researcher, Reflection AI)

---

Overview

This episode centers on the state-of-the-art in artificial intelligence—especially the ongoing quest for artificial general intelligence (AGI)—with expert insight from Jeremy Berman of Reflection AI. The panel explores the anatomy of large language model training, the significance of the ARC AGI benchmark, the evolving landscape of open-weight frontier models, and the philosophical as well as practical implications of "spiky" vs. general AI. With typical TWiT charm and humor, the hosts and guest dig deep into technical and societal questions, highlight the competitive landscape, and reflect on both risks and massive opportunities ahead.

---

Key Discussion Points & Insights

1. Understanding Pre-training vs. Post-training in AI

• [02:51] Jeremy Berman lays out the differences:
  • Pre-training: Consists of training neural networks on vast Internet datasets, teaching them to predict the next word in any sample of data—effectively compressing the “intelligence” of the internet. Pretrained models excel at document completion but aren’t useful as assistants out-of-the-box.
  • Post-training: The phase where models are made useful for human tasks, fine-tuned for behavior like helpfulness or coding, and given personalities by training on curated datasets.
  • Reinforcement Learning (RL): Considered part of post-training, this involves the model training itself on its own generated outputs, iteratively improving its performance and reasoning abilities.

Quote:
"Pre-training is the process of stuffing basically the Internet into a deep neural network ... But the problem is these models are not useful ... Post-training is the process of making that useful for humans and tasks."
— Jeremy Berman [03:06]

---

2. Frontiers and Benchmarks: Reflection AI and the ARC AGI Test

• [09:53] Berman explains his recent achievement: achieving top scores in the ARC AGI test, an “IQ test for machines” designed by François Chollet. While children easily ace these pattern recognition puzzles, LLMs historically performed poorly (4–5%).
• Reinforcement learning dramatically changed the landscape, as OpenAI’s O1 model “smoked” Berman's high score using these techniques.
• Reflection AI’s Approach: The company leverages expertise from DeepMind’s reinforcement learning (RL) (e.g., AlphaGo) and applies RL techniques to language models for more human-like, generalizable reasoning.

Quote:
"What changed the game completely is ... let’s teach these models to do it from scratch ... This is the power of letting the models think for themselves."
— Jeremy Berman [14:18]

---

3. The “Spiky” AI Debate: Specialization vs. Generalization

• [17:30] The panel examines whether the march towards AGI is distracted by the proven excellence of narrow, “spiky” superintelligences (e.g., AlphaGo), or whether the real challenge/goal lies in building systems with generalized, transferable reasoning.
• Berman believes it’s likely we’ll achieve “spiky superintelligence” soon, but is optimistic about integrating true generality—reasoning that transfers across domains, covering "dead reasoning zones."

Quote:
"There’s one world where okay, let’s just build a dataset for literally everything ... but the right answer is we need to build the right environments and the right training paradigm such that the models internalize reasoning for all domains in a general way."
— Jeremy Berman [20:54]

---

4. Inside Reflection AI: Open-Weight Frontier Models

• [28:03] Reflection AI aims to deliver open-weight models at the frontier, unlike incumbents (OpenAI, Anthropic) whose models are closed-access. This serves enterprise needs (e.g., privacy for governments, regulatory compliance) and promotes research accessibility.
• Chinese open-weight models currently lead the market, but Reflection sees a gap in non-Chinese, cutting-edge, open models.

---

5. Training Paradigms: Reinforcement, Distillation, and Research Culture

• [21:31] Berman describes lively lunchroom debates at Reflection AI, referencing the recent “on-policy distillation” paper—teacher-student RL techniques making models learn more efficiently.
• Important internal disagreements revolve around whether models must learn from their own traces, or expert demonstrations suffice.

---

6. Current Limitations & Promising Directions

• [29:28] AGI remains elusive because AIs continue to struggle with consistent reasoning across domains ("dead reasoning zones").
• Current models store knowledge as a web, lacking the "coherent causal relationship" that reinforcement learning can impart.
• Berman speculates that future breakthroughs will require new training paradigms—perhaps with broader, more creative post-training phases.

---

7. The Scale Debate and AGI Outlook

• [35:21] The panel discusses whether continued scaling and LLM architectures will bring AGI, or whether a paradigm shift is needed (as argued by Yann LeCun and others).
• Berman maintains that scaling, especially focused on post-training and reasoning, is promising. He’s confident we’ll solve spiky superintelligence and cautiously optimistic about general reasoning “within the next 10 years.”

---

8. Risks, Openness, and Social Implications

• [39:05] AGI could be as dangerous as nuclear weapons if universally accessible. The potential for misuse is real—“everyone on the planet has basically an Einstein in their pocket” ([39:13]).
• Models from Reflection AI are not yet public but the company is committed to openness.

---

Notable Quotes & Moments

• On the leap to AGI:

  “We will know we've achieved AGI when we can't create tasks that are easy for humans but hard for AI.”
  — Leo Laporte [29:28]

• On the “AI bubble” and cultural moments:

  “I look down. I hold up a bubble. The bubble's filling up the screen. The words AI Are in it in a bubble. ... Suddenly Freddy Krueger s cams come and pop the bubble.”
  — Paris Martineau (reacting to an on-air video gag) [73:50]

• On spiky superintelligence vs. true generality:

  “I am very confident ... we will build spiky superintelligence, ... I think it’s more likely than not that in the next 10 years we have new ideas that will lead to true general reasoning.”
  — Jeremy Berman [33:29]

• On the “lunchroom culture” at Reflection:

  “People at Reflection are very smart ... I’ll toss out a few theories. They’ll tell me why I'm wrong. I'll tell them why they're wrong ... It’s paper-sharing and saying, ‘See, told you this is OK...’”
  — Jeremy Berman [21:31]

• On open-weight models for the community:

  “It’s really great to be able to contribute to the community ... If we’re able to build a great model, we can give it to researchers ... for scientific discovery.”
  — Jeremy Berman [28:03]

• On AGI safety (and jokes about doomsaying):

  “You’re not with what’s his [name] ... Yudkowsky says if we get there, we’re all dead?”
  — Leo Laporte [38:56]
  “It’s reasonable to assume they’ll be about as dangerous as nuclear weapons ... but not more than that.”
  — Jeremy Berman [39:11]

---

Timestamps for Important Segments

• [02:51] — Pre-training vs. Post-training, RL explained
• [09:53] — Berman’s path to AI; ARC AGI test and breakthroughs
• [14:18] — AlphaGo’s reinforcement learning as paradigm inspiration
• [17:30] — Spiky vs. general AI and field-wide debates
• [21:31] — Reflection’s research culture, lunchroom debates, on-policy distillation
• [28:03] — Open-weight models: technical, business, and philosophical rationale
• [29:28] — Consistent reasoning, “dead zones,” challenges ahead
• [33:29] — AGI outlook: spiky superintelligence, forecast for general reasoning
• [39:05] — Societal risks of AGI and comparison to nuclear arms
• [73:50] — Podcast in-joke: “AI bubble” video bit with Paris
• [81:23] — Paris' Halloween as Log Lady, whimsical tangents (charm of TWiT)
• [129:03] — Data centers in space, practical AI applications
• [135:03] — Favorite AI tools (Claude, ChatGPT, Perplexity), research best practices

---

Conclusion

This episode is a goldmine for anyone following AGI progress, LLM architectures, or the interplay between open research and real-world applications. Jeremy Berman’s grounded, lucid explanations offer a roadmap to where AI is today and where it could go—demystifying "post-training," the importance of RL, and the choppy waters between specialized and general intelligence. The conversation is peppered with insight about the industry, caution about risks, and a genuine sense of curiosity and excitement about the possibilities ahead.

---

Additional Tidbits

• Paris is heading to the American Museum of Tort Law—one of many quirky cultural asides.
• The open-source, researcher-friendly ethos remains strong at Reflection AI, paralleling broader calls for transparency in AI systems.
• The co-hosts digress into tangents on AI-generated ads, copyright debates, and even vintage pizza—showcasing the podcast’s signature blend of depth, warmth, and humor.

---

Further Listening/Reading

• Jeremy Berman’s Substack (for technical breakdowns and ARC AGI solutions)
• ARC AGI benchmark (Francois Chollet)
• OpenWeight models and the shifting enterprise AI landscape

---

Closing

"Thank you, Jeremy, for helping us... You're right in the middle of the most exciting thing to happen, I think, in human life."
— Leo Laporte [40:10]

---

For a detailed, accessible dive into the most important issues surrounding AGI, this episode of Intelligent Machines delivers expertise, clarity, and plenty of memorable moments.

Transcript

[A]

It's time for Intelligent Machines. Parris and Jeff are here. Our guest this hour, Jeremy Berman. He is a post training researcher at Reflection AI and his AI recently just got the highest score on the ARC AGI test. How did he do it? What is post training and what's the future of AGI? All coming up next on Intelligent Machines. Podcasts you love from people you trust. This is twit. This is Intelligent Machines with Paris Martineau and Jeff Jarvis. Episode 844, recorded Wednesday, November 5, 2025. Poob has it for you. It's time for Intelligent Machines, the show. We cover the latest in artificial intelligence and robotics and the smart little doohickeys surrounding us all. Look, I've got smart doohickeys everywhere in the house. Jeff Jarvis is here, professor emeritus of journalistic innovation at the Craig Norma Graduate School of.

[B]

Craig.

[C]

You almost had it, Leah.

[A]

I do. We made a deal that if I got to the end that we wouldn't play the jingle. Jeff is the author of the Gutenberg Parenthesis and magazine, a history of magazines. And when does hot type come out? Because I'm loving it. June. I have to wait till June.

[B]

Oh, but tell me if I put anything up. Tell me if I got anything wrong.

[C]

It's got to be warm weather for hot type. Steph's going to have a beachside launch party. Right.

[A]

That word Smithing comes to you from Paris Martineau, who is a investigative journalist at Consumer Reports. Hello, Paris. Bonjour, Paris. So we have a very interesting guest joining us for as we often do the beginning of the show, we like to do interviews. You may remember when Deep Seek kind of changed the world with its inexpensive model that was. Was kind of using new technologies to create better LLMs using reinforcement learning. Well, there is a company called Reflection AI that has just raised $2 billion to challenge Deep Seek, founded by two former DeepMind researchers. And we've got one of them with us today. It's great to have Jeremy Berman here. He's a post training researcher at Reflection AI. Jeremy, welcome.

[D]

Thank you. I'm not one of the founders.

[A]

Oh, you're not one of the founders. You're just a. A guy.

[B]

That's the next company he does.

[A]

That's the next one. Oh, that's right.

[D]

Someone they pulled in.

[A]

Okay, good. So tell us a little bit about what post training research is.

[D]

Sure. Yeah. So I guess to start we should talk about what pre training research is. Okay.

[A]

Pre training, that makes sense.

[D]

Yes. Basically what all language models that were used were doing. Maybe pre 2024 mid 2024. And the idea is we had this incredible breakthrough which is we can stuff the entire Internet into these deep neural networks and we can basically with, with the goal of the neural network mimicking the Internet. So it's like a big Internet document, right? So we train these deep neural networks to predict the next word that it sees on the Internet and eventually it gets really good at doing that. And you can imagine a system that's really good at predicting the next word in a sentence has to have some sort of imbued intelligence to be able to do that. And so pre training is the process of stuffing basically the Internet into a deep neural network and imbuing the intelligence of the Internet into that network. Some consider it kind of like compressing the intelligence of the Internet into a model. But the problem is these models are actually not useful. So you have a pre trained model and it's just a document completer. So it just goes next word, next word, next word, as if it's just building a webpage on the Internet. So that's pre training. And then post training is the process of making that useful for humans and tasks. And so as an example, when you talk to ChatGPT, ChatGPT is a post trained model in that it knows what a user is, it knows what an assistant is, it's trained to be helpful. These are not things that come out of the box. It's an out of the box language model is just it completes documents. And so post training is then the process of turning that to be useful for certain tasks. And I would say post training, maybe from 2023 to 2024, 2025 was making it useful. So this is how you should respond. You should respond in this way we can imbue a personality and this is all by showing it examples of what we want.

[A]

And then in some respects this is what differentiates anthropic to OpenAI to Google. Is, is this post training? Right. Because essentially the models are all trained on roughly the same stuff.

[D]

That was actually my assumption before I really got into the weeds. And we're doing pre training and post training. At reflection, that is the prevailing wisdom. I think that's actually not true.

[A]

Oh, interesting.

[D]

The labs have actually quite different ways they pre train their models. The general structure is the same, but the data they're training on, the mixtures of data they're training on is actually quite different. For example, some models include certain types of code snippets, some models include certain types of data. As an example, I Think Gemini has grown into the incredible model it is today because of an incredible pre training effort at Google, which I believe is probably the strongest in the world right now. But yes, I think your general principle holds, which is post training is where you give the model the personality and it helps to have a very strong pre trained model. But post training is yes, where anthropic makes their models very good at coding, where OpenAI can make their models very good at reasoning. Yeah, these are things mostly done in post training.

[A]

So as I understand it, we've talked a lot about this on the show. You have a corpus of information, common crawl or a giant dump of pirated books or the Internet as a whole and you, you, you, you tokenize it and then you also there's some fine tuning that goes on in the to after the tokenization. Right. Even some reinforcement learning. Is that what you're talking about with post training or is it even after that?

[B]

That's still pre.

[D]

So I think actually the line is a bit blurry. It's really. And actually at reflection we called it training because a lot of post training, Post pre, whatever.

[A]

It's training.

[D]

Yeah, right. And I think maybe you can think about it like what is the objective of the neural network that you're trying what is the objective that you're trying to teach the neural network? So in pre training that objective is predict the next token. And there are certain post, like if you want to train it to be a helpful assistant, you generally have a data set of helpful user assistant messages.

[A]

Right.

[D]

And then you can post train it to predict the next token. But then there's a whole separate area which is called reinforcement learning, which is fundamentally letting the model learn for itself, reinforcing training the model on its own traces. And this is what Deep Seq R1 did. It brought this whole field into the next generation of actually having the model learn for itself. And so that is also bucketed in post training. So I think generally the lines are a bit blurry, but it's helpful to think of what is the objective here. So the objective in pre training and some of post training is next token prediction. Almost like memorizing the data set, imbuing the data set, and then there's another set which is reinforcement learning or on policy learning, which is I'm going to have this model generate a bunch of answers, I'm going to take the best ones and then I'm going to feed it back into the training corpus until the model learns to basically builds its own reasoning and thinking circuits and you don't get typically that type of generalization and thinking for yourself when the objective is just predicting the next token. You really want the model to be generating thoughts on its own and then reinforcing on those good thoughts.

[A]

I want to take a step back here because your history is kind of interesting. You do have a master's in computer science, but you were an entrepreneur. You had two startups, part of Y Combinator's winter batch in 2019, and then said, you know, I kind of like. I kind of like this research side of this. I want to get more into the. The AI research size side. You read Jeff Hawkins book? I've interviewed. We've interviewed Jeff Hawkins. In fact, we got to get him on the show. He's fascinating guy. He's the guy who came up with graffiti for Palm and is an. Is a neuroscientist. And he's written some really interesting books about how humans think, how our brains work compared to how machine brains work. And I, according to our research, that's how that kind of inspired you to. To get into this research side on this. That's a.

[D]

That.

[A]

So you're relatively new to this?

[D]

Yes, I would say I've only been in this field for about a year and two months.

[B]

Wow.

[D]

Really?

[C]

A year and two months.

[A]

I mean, this is a new field, so it's not the end of the world, but that's interesting.

[D]

Yeah, yeah, I. So I was. I had a startup. It's doing well. We have, you know, we have an office. Still around still. Yeah, still around. And I read Jeff's book and I could just get a sense that if we are able to figure out general intelligence, artificial general intelligence, it will be the most important thing that we do.

[A]

So you're working pro, you're working on AGI more than anything else.

[D]

Yes.

[A]

This is your inspiration. And in fact, the reason you're here is because you took on the Arc AGI Challenge and achieved a remarkable goal. First place in 2024 and then broke.

[C]

Records again this year.

[A]

Yeah, 76%, 79%, 79.6%, which is remarkable.

[C]

Can you tell us a little bit about what the ARC AGI Challenge is and what you did in these cases?

[D]

Yeah. So ARC AGI is like an IQ test for machines. And it's so simple that children can get 85 to 100% on the V1. But language models at the time, and this was 2024, late 2024, were getting like 4%, 5%. These were the best language models we had. And that was what really sparked my interest because. And let me. So what it is is it's a puzzle. You have grids, input and output grids, and there's a common transformation rule that you can apply to the input grids to get to the output grids. Again, simple children's puzzles. And the key is, can you, given an input, generate the output grid? And children can do it. Everyone on this panel will get 100% on it. But the best language models in the world in 2024 were getting 4, 5, 6% totally not saturated.

[C]

Why do you think that is?

[D]

So the reason was they were pre trained and then post trained, but not reinforcement learned. And so these models are kind of, you can think of them like stochastic parrots, which is the objective of just predicting the next token and that it's soaked up all of the knowledge of the Internet and then on some post training data sets, but it hasn't learned to fundamentally think for itself. And ARC AGI requires you to be able to think on your feet because you've never seen these challenges before. These are not on the Internet. You can't hack this test. You have to be able to generate on the fly thinking. And this was like, this is a beautiful test. Francois Chollet created it in 2019 and it has really stood the test of time.

[A]

But then he's one of the founders of Zapier, by the way. And still. And still.

[D]

Mike Knoop.

[A]

Oh, I'm sorry.

[D]

That's okay. He's the other, the other guy.

[A]

Okay. And I'm taking, by the way, the. You have daily. They have daily puzzles on the ARC website. I'm taking the test right now and not, not doing well. Let's see.

[C]

Do you think that the LLMs also host a PUZ podcast while they do this as well?

[A]

Yeah, right, exactly.

[D]

That looks like a V2 puzzle, which is harder. And I'll get to that.

[B]

Oh, you're just giving them an easy out.

[A]

But you're right. These are simple. But this is. These are not. By the way, one of the things, this is not verbal. And one of the things that the AIs, the LLMs we're used to are very much about is verbal.

[D]

But it's important to note that these are the, these are non verbal but easy. And so this goes to Francois Thesis, which I agree with at the time. Which I agreed with at the time and which is if language models are not trained on a thing, if it's out of their distribution, their training distribution, they will not do well.

[A]

Right.

[D]

And this is the missing generality, right? Like humans are very good at on the fly thinking, learning new skills, learning a new board game, learning a new puzzle. Language models at the time were not because they were trained to soak up information, they were not trained to think for themselves fundamentally. And this. So I was able to get top score in 2024 by basically generating a ton of Python programs. I had a language model generate a ton of Python programs and then I would have basically a Python executor test those programs. And then the most promising programs would then be put in a revision loop where I would have a language model update incrementally evolve the programs to be better and better. But what's important about that is it got first place. And then two weeks after that, 01 came out. OpenAI's O1 model, which is the first model which, using the Deep SEQ paradigm, which was taught via reinforcement learning to fundamentally think for itself. And it smoked my score. I was first for two weeks and then it totally smoked my score. And this like totally lit a light bulb. You know, I had a light bulb moment where I was like, this is really the new paradigm here. We have cracked the code for how to teach basically to bring the general into distribution. And that is through reinforcement learning, through having the model try out different answers and then taking the best ones and then reinforcing on those and doing that in a loop.

[B]

So what sets Reflection AI apart?

[D]

So given this, the founders of reflection were at DeepMind when DeepMind was really pushing forward the reinforcement learning paradigm. And the best example is AlphaGo. So AlphaGo is a program that via reinforcement learning. So first actually let's take AlphaGo V1. They took human, the best humans, they took games of the best humans and then trained the model, pre trained the model basically to soak up the information from those games and then play Go. And this worked really well. And they got to human level performance. But what changed the game completely is that then they said, you know what, screw the human data, let's teach these models to do it from scratch. And then we're going to reinforcement learn on their good ideas. Have it basically self play, explore the field, use no human traces and see what happens. And this is what led to AlphaGo handily surpassing the best human. And what is often cited is move 37, which was a move that humans at the time thought was crazy, but turned out to be this brilliant move that no human had really thought of to play. And this is the power of actually letting the models think for themselves. And so at Reflection. We have language models that are generally, we have this reinforcement learning paradigm which we know produces superhuman intelligence. So why don't we combine them? And that is kind of the principle, I would say, that reflection has been founded on, which is extreme expertise in AlphaGo style reinforcement learning, plus expertise in reinforcement, sorry, expertise in language models. And that is kind of our origin story.

[B]

Jeremy I read an interesting paper from a colleague at Stony Brook University two weeks ago that I'll make this as quick as I can. They took given authors and they had writers summarize their work, and then they had AI summarize their work, and then they judged that against two pools of people, one who were expert and one who were just plain folk. And for the AI was fine for the plain folk, but the experts threw up their nose. Then they trained the models on the entire OOV of the authors, and then the AI summaries beat the human writer summaries with everybody, including the experts. And this really struck me, it struck me as a writer, that I think that there's opportunities in publishing that we can't see in terms of creating a Jeff Bottom but it also struck me about the specialization of that. And so what I want to drill on a little bit is this notion of general intelligence and the effort at that versus what we know is the case is that AI can do specific tasks like AlphaGo really frigging well. And so I wonder, is the seeking general a distraction from all kinds of new frontiers that could be met at a superhuman level, but not a general level? Did I make any sense?

[D]

Yes, this is currently, I would say, the question of the day, which is, are we building spiky superintelligences Right, where we have this data set, it's clean and we know for a fact, it's just a fact, that if you have a distribution of data, deep neural networks can learn that distribution and do it basically as good as we need it to. Right. But the problem is you can almost imagine it like it's overfit to that training data. So if then you give it data that it hasn't seen before that's sufficiently different, it will do less well. And so this is the spiky. This is the spiky AI paradigm. And then the other paradigm is we figure out how to build generality, how to build basically the skill that builds other skills into these machines.

[C]

And.

[D]

And that you would probably see as both superhuman and superhuman at everything. And the latter is what I'm most interested in. But I think people at reflection are mixed. People generally are mixed, it's not clear which one will win. I have a theory for why I think that we are going to be able to inject generality into these models, but it's not certain.

[A]

You may remember I referred you to a video which you never watched about how you can take an overfit model, right, reduce the parameters and get it to actually generate an algorithm instead of trying to brute force the solution.

[C]

Leo, is this the seven hour YouTube video you.

[A]

This is a shorter one but it was a very interesting theory. And if it's true, if you could have a AI generate an algorithm for solving as opposed to a simple well 2 plus 2 equals 4 then 2 plus 3 equals 5. Have a, have a truth table that would be much closer to human intell. An actual intelligence. Is that this is the kind of thing we're talking about, right?

[B]

Tell us about your theory.

[A]

What's fascinating is that there's a lot of room for innovation here, right? There's a lot of places you can improve the training, the post training.

[D]

Yes, I agree and I think at the end of the day reasoning with sufficient creativity is the engine of generality. And I think right now what we're doing is we have these data sets that are very particular to a certain set. So we have a math data set and we can train it to be very good at math and we can train it to be very good at reasoning and math. The problem is it's almost like the reasoning and math circuits that we've trained together are fused in that then we ask it a question about something else and it generalizes a little bit, but not as much as we would hope it to. And I think this is, you know, what I'm most interested in is figuring out like why exactly that is happening. It's almost like when you ask a language model, a task that it hasn't been trained on, it's like an adversarial attack because you're hitting weights that it's memorized in pre training. This is actually what happens in ARC. You have these PhD level machines that are making obvious blunders on again questions that are very simple and it's because it's hitting weights that were pre trained and when it hits these pre trained weights it has not learned to think for itself with this, with these circuits. And so it ends up guessing. This is where it hallucinates. It just goes down weird paths. And I think this is, you know, there's one world where okay then let's just build a data set for literally everything but Then, you know, when we have new data and the world is full of new data. Right. Our brains are reasoning engines for a reason. Probably evolution knew that we're always going to encounter new things. And so I think the right answer is we need to build the right environments and build the right training paradigm such that the models internalize reasoning for all domains in a general way.

[B]

I want to hear about the lunchroom. You said there's disagreement even within reflection. Yes, I want to hear about those lunchroom conversations.

[A]

You want that, by the way, you don't want this kind of everybody saying, oh, yeah, we know how to do it.

[B]

That's a very theoretical level at which to express disagreement. How does that, how does that discussion go?

[D]

Well, the people reflection are very smart. It's, it's. It's very fun. Lunches are very fun. We always have lunch together. And yeah, it's very fun. How does it go? Well, you know, I think I'll toss out a few theories. They'll tell me why I'm wrong, I'll tell them why I think I'm right, and then we'll go from there. I think, yeah, it's fun. What is it? It's probably not. I think it's a lot of paper sharing and saying, look, I told you this is okay. I mean, I think there's a really good paper, it's technical, from thinking machines that just came out called on policy distillation, which is what they did is basically they took a language model, they have a teacher model and a student model. And usually the way reinforcement learning works is you have the model try and, let's say, answer a math problem 100 times, you hope that one of those 100 is correct, and then you can take the reasoning that it used for that correct one, and you can train it on its own thinking process. But that's very inefficient. And also, what if the model doesn't even get to the right answer? Right. You've just wasted 100 traces. Well, on policy distillation is a way for the teacher to actually, for each next token, for each next word, tell whether, oh, I think this was a good next word. I think this was a bad next word. So you get a very dense signal, and so you don't have to do as many rollouts, and the student model learns quicker. This is just something that came up in a lunch discussion recently, which I think is interesting because there's another debate which is, do they even need to learn from their own traces, or is it okay that we have expert traces and they can just soak up the expert traces and they all of a sudden can kind of imbue the intelligence from the expert. I'm not in that camp. I think fundamentally we need to reinforce its own circuits. But yeah, that's kind of a. I would say that's a window into our lunch conversation.

[B]

So then what you want to say, I'm going to show.

[A]

Hold on just a second. We're talking to Jeremy Berman. He is a researcher and a very interesting person at Reflection AI post training research. And you don't win the ARC AGI benchmark, but you did very well. In fact, the new benchmark results come out in a few weeks. Do you know how you did yet?

[D]

Well, my model is currently at the top with 20 with on arc v1, 79 and then arc v2, almost 30. I think that's going to stand as the top score. But we will see.

[A]

We'll find out in a few weeks. I just want to point a trophy.

[D]

You do get a trophy and instead of a trophy you get a lot of cash. If you're able to get above a certain threshold, let's say I think it's something like 75 or 85% on RV2 under a certain budget. And it has to run on the Kaggle on the notebook. That's a small amount of interest.

[A]

So you do this on Kaggle. That's interesting. And you have a budget for how much you can a 50 time or.

[C]

Money budget for 120 evaluation tasks.

[A]

By the way, I have failed miserably at V2.

[D]

V2 is harder.

[A]

Well, my problem is. I understand. So these are the examples and this is then what you're supposed to do. And you're supposed to see what the transforms are. And it should be very easy. But this one is harder because the transform looks like it's a certain number of steps in color. And I just can't quite figure out maybe this wouldn't be good if you were colorblind, let's put it that way. So yeah, I've tried twice and failed twice.

[D]

I also just one distinction is that I'm using models that are. I'm using language models via their API.

[A]

Interesting.

[D]

And so there are two separate ways to do it. There's a way that you run your experiment on the Kaggle notebook, which is extremely small in compute. And then there's a way that I do it, which is to use an API for the models for other models. And those are different. So I'm on the public leaderboard, technically and what I'm describing on Kaggle notebooks is called the private leaderboard, and they're the ones that are eligible for the prize. What I'm doing is not actually eligible for the prize, But I get $10,000 of compute per run and my goal is really to just see where the frontier is. That's my biggest goal, is just what are the best language models capable of?

[A]

You do this in Python?

[D]

Yes, well, the program is written in Python, but my Arc v2 solution, which I submitted before coming to Reflection, actually, instead of generating Python programs, I generate the programs in natural language, so they're actually just bullet points. And then I have sub agents or sub instances verify whether the plain English instructions actually work on the grids.

[A]

That's cool.

[D]

Is basically taking advantage of the fact that thinking models exist now. They didn't when I was originally doing this. And so with thinking models, they're actually much better at reasoning on these types of things.

[A]

So you're vibe coding your solution, in effect.

[D]

Yes, that's a good way to think about it.

[A]

I'm sorry, I didn't mean to interrupt you, Jeff. I just wanted to reintroduce Jeremy for people who joined.

[B]

Yeah, that's good. So. So when you have that lunchtime discussion and you say aha, no, I'm going to show you. You need compute to do that. I'm guessing that allocation of compute resources is the key to be able to try these things. What's the currency that you then work in? Or does the company say, this is what we need to try next? And you're trying to convince the company that that's what the company should do? Or are there five different experiments going on? How does innovation go on in a company like this?

[D]

So I guess I can speak at a high level. I can't talk about, I guess, the specifics, but I would actually, taking a step back, the goal of Reflection right now is to build the best or a frontier open weight model. And so our number one goal is actually to just make sure that we are doing the things that we know to work really well, basically, you know, excellence in execution. And so that's our number one goal. And it's of course great to do research and we do do research. And I think the research we're doing is really interesting. But our number one goal is making sure that we are able to produce a model that is really, really good and open weight.

[B]

So talk about the open weight part. That also separates you from some of the others. There's a mission there for our Audience, why don't you talk about that just a minute.

[D]

Yeah. So right now the frontier language models are closed, which means that if I wanted to run them, I have a business, I can't just download the model and then run it on my own computers. I have to send my request to OpenAI or Anthropic and from a business use case. This is challenging for private workloads or for regulatory reasons. And so if I'm a government, right, I want to use my own models on my own hardware, then I can't use frontier language models. And also there are various regulations or reasons why I wouldn't want to use Chinese models. And the Chinese models are currently overwhelmingly at the top, the best models in the world for open weight models. And again, open weight means you can download them, I can click a button, download them to my computer and actually run them on my own computers. And so we've found this gap in the market where there are a lot of enterprises that really want to run their own models for various reasons, but they find themselves, if they do want to do this, they're using old models or they're using models that are not the frontier. So we're really going after that segment. And then there's a second proponent, which is just philosophical, which is it's really great to be able to contribute to the community. If we're able to build a great model, we can give it to researchers and researchers will be able to run experiments, they'll be able to fine tune it for scientific discovery, for things that they are interested in. And so I think from a business perspective it's interesting. But then also for just a research perspective, it's inspiring.

[A]

You say we will know we've achieved AGI when we can't create tasks that are easy for humans but hard for AI. Tasks like this arc AGI puzzle, which turns out to be hard for a human too. But that's another, that's just me. But more importantly, I think, and I'm reading your substack where you talk about how you solve this, humans are able to extend reasoning from one discipline to another. We have consistent, you say consistent reasoning that transfers across domains. So an economist who's good at logic can become a programmer when they learn to code because it's a reasoning domain that's the same. It's not the same knowledge domain, but it's a reasoning domain. You say that this is something that AI is currently aren't good at. They have dead reasoning zones, they cannot cross disciplines that way. And is is the Training the solution to this or. And also, have I misstated what you were saying?

[D]

I would say just generally they, no, no, I, I, it's for like OpenAI's models are very generally good. But that's because OpenAI has really great training processes and they have, they fill.

[A]

In the dead zones. They know where the dead zones and they fill them in. But we want to get a solution that automatically that doesn't create dead zones.

[D]

Right, I agree. And that is what I'm most interested in doing. And I think the way you can think about it is we pre train these models and knowledge from pre training is stored like a knowledge web. Right. Where there's not a cohesive or coherent causal relationship between things directly in that you haven't tried out your ideas. If you're a language model only pre trained in the real world. Right. You haven't gotten feedback on how your brain works. You've just been soaking up the information. So you have this what I call a knowledge web where you have semantic understandings of things. Right. You can connect two things, but you are missing this causal understanding of the world because you just haven't tested your ideas out in the real world. And so I think what reinforcement learning does is it slowly shifts knowledge from this web to more of a cohesive knowledge tree. And I think the dead zones are places where we haven't yet found, we haven't yet merged or molded these weights to be coherent in the real world. And I think this might come from just a different training paradigm altogether where we're able to maybe shrink pre training, massively extend the scale of post training or some other ideas that I'm not exactly allowed to speak about right now. Right.

[A]

But that's what we're doing. Right. We're trying stuff until we figure it out. But what's good is that we now have. It sounds like we have a goal. We haven't, we have a. At least we're not sure if this is going to work, but we have a pretty good idea of what we want to accomplish and now we can try different things to try to accomplish it.

[D]

Yes. And I think this is also why it's important that we have a lot of open weight models that are strong because it's very useful to be able to get in there and actually try out your ideas.

[A]

Yeah. If you want to see what his English language solution looks like. Jeremy on his substack has published a few examples. It's kind of amazing really that you can write a prompt like this to solve something like Arc AGI.

[D]

Yes. Yeah, it's. Models have gone a long way and they will continue. They will never be worse than they are now.

[A]

Yeah, I think it's exciting time and, you know, we don't. What we don't know is whether we'll solve this. Right. I mean, it's a terra incognita. We've not been here before.

[D]

Yes.

[A]

Do you feel fairly confident, though, that there is an answer, there's a solution out there and that we can teach machines to reason?

[D]

I think I'm very confident that we will be able to build spiky superintelligence. We have a data set. We can train the model to be extremely good at that data set. I think it's more likely than not that in the next 10 years we have new ideas that will lead to true general reasoning that spiky is not the solution.

[A]

Spiky means that there's edge cases where it's just going to hallucinate. It's just not going to give you the good answer. You want it to be more continuous. You want. Yeah.

[D]

Yes. And a good example is Ark v3, which is now actually a video game. And there are levels and it's. Yeah, it's like an Atari game, but each level adds a new objective, you know, a new twist. And the best language models, frontier language models, cannot do it at all. And it's fun. You can play a game yourself. Over time you'll learn and you'll. You'll understand how it works. The best language models totally fail at this. And I think this is a good distillation of what I mean by we definitely do not have general intelligence. If language models are struggling to even pass the first level. I don't think a language model is even past the first level of a real architecture.

[A]

Leo's struggling to pass the first level.

[B]

So Yann Lecun says that language models have kind of hit their apotheosis and that he's one of those who looks at real world models and other things as the next paradigm. And then we had Karen Howe on her book about Empire of AI a few weeks ago, and she was talking about the controversy of those who think that scale is the answer and those who think that no smarter training and different paradigms are the answer. Just in terms of just at a high level philosophical view. Are you LLM side still or got to do something new? Are you scale side still or something else? Where do you see that future building?

[D]

I am, I am more. I am more confident than not that language models will get us there. And I am also confident that as we continue to scale the systems will be smarter and smarter. But I think it's important that we're not necessarily scaling and pre training but that it's post training where we're scaling. I think generally models that are bigger are better as long as they're trained appropriately. But I'm actually not exactly sure if that's necessary. I think it's very possible that we will figure out reasoning and the models will be no bigger than they are today. Or maybe we'll figure it out and then the models will be able to build models that are much more efficient that are the size that they are today. I don't think there's any physical reason why they can't be.

[A]

This is the problem though that I have with Yann Lecun and Karen how and they're in different categories but in general with the speculation of well that's probably not going to work. We don't know, right? This is what's.

[C]

We don't know that it's going to work either.

[A]

We don't know it's going to work.

[B]

We don't know it's an investment risk.

[A]

And but we, we gotta try every avenue and have the smartest people like Jeremy thinking I love the lunch, I would love to be in that lunchroom to get these smart people together thinking about it and trying stuff and until you, you know, you gotta try it all and see what works and see what doesn't work, we just don't know. So speculation from Yann Lecun that oh LLMs have come to the end of the. It's just, it's speculative, it's we don't know.

[D]

Here's maybe I can steel man him and I actually want to be careful because I'm sure I'm not going to do a good job. So maybe let me steel man an argument that I think does make sense for people that have that perspective, which is fundamentally neural networks learn the distribution they are trained on, which means you give it data and it will learn that distribution but it will not be able to generalize outside of that distribution. So that is like Francois generally says this right? Or maybe a variant of this and this is true, but I actually don't think it's very hard to fit all of reasoning into that body and all you have to do is fit reasoning into distribution and that is it. Because reasoning is the engine that builds all knowledge. You can deduce from simple axioms, you can deduce almost everything and so it doesn't need to be in distribution if it's deducible through coherent reasoning. And so if we have coherent reasoning and the ability to have creativity and taste on what to reason about, I don't see why it is the case that language models will not be able to discover new things. And discovering new things is to me the heart of general intelligence.

[B]

That's the. Yeah.

[A]

Love it.

[B]

So talking to us is probably like the question about to ask what's it like when you go and have Thanksgiving with the family? They say what are you doing, Jeremy? How. How detailed do you try?

[A]

You're still doing that little computer thing that you were doing.

[B]

Or maybe it's just like talking to us right now. I don't know.

[A]

It's probably very similar.

[D]

Yeah, no, no. What happens is, you know, my mom is like, you really need to clear your plate. And I'm like mom, I'm working on AGI. Maybe someone that's. Maybe, maybe the lawyers can clean my plate you're doing.

[A]

So I think I understand completely why you said I'm gonna do this because it is probably the most interesting and exciting and thing that humans are doing right now. And yes, it may be a dead end. We don't know. But you gotta try and if you get there. Well, you're not worried about we're all be dead. Are you? You're not a. You're not.

[D]

No, no, I'm not.

[A]

You don't not. You're not with what's his Eliezer Ykowski says if we get there, we're all dead. No. Okay, good. All right.

[D]

It's reasonable to assume that they will be around as dangerous as nuclear weapons and that that's pretty dangerous.

[A]

That's more dangerous than I would have thought.

[D]

Okay, well you know you're giving. If we are able to make these extreme. If we are able to do what I'm describing right. Actually build general intelligence that runs in a reasonable efficiency. This means that everyone on the planet has basically an Einstein in their pocket. Right. So obviously there's more to building a bomb than just Einstein's brain. But there's not much more to a lot of dangerous things if you have, you know, an army of Einstein's working for you for not so expensive, you know, for. For you know, maybe $100,000. So I think it's on the order of risk as nuclear weapons. But I don't think it is more than that.

[A]

Has. Has the reflection put any models up on a hugging face or anywhere for us to play with or these are all internal still.

[D]

No, they're internal, but we. I don't want to say. Any time.

[A]

It's a goal.

[D]

It's a goal.

[A]

Well, yes. Reflection. AI building frontier open intelligence. Imagine going to lunch with the smartest people in the world.

[B]

Jobs open there, folks.

[A]

Yeah. How is it that we reason? What is it that we do? And you're right in the middle of the most exciting thing to happen, I think, in human life. It's great to have you, Jeremy. Thank you. Thank you for helping us.

[B]

Thanks, Jeremy.

[A]

Neanderthals understand what Thanksgiving Day table. Yes.

[D]

Thank you very much for having me. And I just. Mom, if you're watching, that was a joke.

[C]

He's gonna clear his plate.

[A]

Your plate, Jeremy.

[D]

I will clean and then we'll have.

[A]

You wash the dishes after. Yeah, I don't know why I'm talking like that. Thank you, Jeremy.

[D]

Appreciate it.

[A]

Reflections here in the room. I know. Because I'm a dad. I know. Thank you, Jeremy. Take care.

[D]

Thank you. See ya.

[A]

We will continue with intelligent machines in just a little bit. Got some AI news. Paris Martineau. Jeff Jarvis. You thinking about a career change now? I am.

[C]

What's the thinking about becoming a philosophy major?

[A]

Yeah, there's no future in podcasting if.

[B]

You can't do the thing that he called simple.

[A]

I can't even do the little.

[B]

I can't. Did you try V1?

[A]

Kids can do.

[B]

Did you try V1?

[A]

No.

[C]

Okay. Well, I think trying to do the thing while you're interviewing someone on a podcast is probably a little different than actually trying it.

[A]

I could do it with one.

[B]

You two were trying to give him an easy out.

[C]

I was.

[A]

Thank you, Paris.

[D]

Thank you.

[A]

All right, we'll have more in just a bit. Our show today, brought to you by Threat Locker. Now, it doesn't take a great brain to understand that we are in a world of hurt and business ransomware is just killing us. But there is a solution. Threat Locker can prevent you from becoming the next victim because it's zero trust. Threat Locker makes zero trust easy. Zero Trust is the way Threat Locker zero trust platform takes a proactive, and this is the key deny by default approach to security. It blocks every unauthorized action. Unless it's explicitly approved, it can't do it. Which protects you from not just from known threats, but from completely unknown threats. Zero days. Things nobody's ever heard of because you didn't give them permission to do anything bad. That's why companies that can't afford to be down for even one minute trust threat locker like JetBlue uses threat locker. The Port of Vancouver uses Threat Locker. Threat Locker shields them and can shield you from zero day exploits and supply chain attacks while providing complete audit trails for compliance. That's a nice side effect of this. You know exactly who did what, when and who couldn't do something. Who was blocked. Right. That's all built into the Zero Trust platform. As more cyber criminals turn to malvertizing, this is something that should scare every business because there's it's almost impossible for you to keep your employees using a browser from clicking a malicious link. Malvertizing means you need more than traditional security tools. Attackers create convincing fake websites. They impersonate popular brands, you know, AI tools we all know or software applications. They promote them through social media ads. They put put high they to hijack accounts and they put it up on, on X and Blue Sky. And then the bad guys use, they're so clever. They use legitimate ad networks, they buy ads to deliver this malware to legitimate sites, sites your employees visit every day. That means anyone browsing on a work system is going to be exposed to this malvertising. And the problem is your traditional security tools often miss these attacks because they use fileless payloads. They run in memory, they exploit trusted services, they bypass typical filters. This is why you need Threat Lockers innovative ring fencing technology. It strengthens endpoint defense by controlling what these applications and scripts can access or execute. If not explicitly permitted, they can't run. That contains potential threats. Even if the malicious ads get to the user on that laptop, in your office, in the lunchroom, even if they get to the device, they can't run, they can't do any harm. Threat Locker works in every industry. It supports PCs and Macs, it provides great support from the US 247 and of course as, as part of this platform you get comprehensive visibility and control. Ask Jack Senisap, Director of IT Infrastructure and Security at Rednor's Markets. He says, quote, when it comes to Threat Locker, the team stands by their product. Threat Locker's onboarding phase was a very good experience. They were very hands on. Threat Locker was able to help me and guide me to where I am in our environment today. Get unprecedented protection quickly, easily and cost effectively with threat locker. Visit threatlocker.com TWIT to get a free 30 day trial and to learn more about how Threat Locker can help mitigate unknown threats and Ensure compliance. That's threatlocker.com TWIT we thank them so much for their support of intelligence machines. Back to the show Jeff Jarvis, Paris Martineau. Paris. You're young. Do you ever think that maybe you give up this journalism thing and start going into AI research?

[C]

Nope. I briefly had a. I. I thought about the hypothetical today because I saw that Anthropic is hiring for two, like, AI writer roles, I guess, for their website with. And some. I found out about this because some journalist was posting a screenshot of this salary range, which is like 225k to 350k. And I was like, that would be nice. But then I was like, would I want to not be a journalist and instead write a corporate blog for Anthropic? And the answer was no.

[A]

I did this when I was about your age, actually.

[C]

I thought, you know, briefly.

[A]

No, I thought to myself, sort of. It was the anthropic of the 90s. I thought to myself, I can't be a deejay for the rest of my life. That's a.

[B]

Okay.

[C]

I'm so sorry. Anthropic of the 90s was working as a deejay?

[A]

No, no, no. I was working as a dj and a guy came to me who was doing a startup, a computing startup, and said, look, why don't you clearly know this stuff? Why don't you come work for me? You could work for a startup. I'll. I'll make you a, you know, investor in the company and we can do it together. Would be. You'd be like, the third or fourth company was. It was called Paracop. It was parallel computing for science. And it was. It was probably 90 or 91 somewhere around there. So this was a. You know, this was like when everything was starting to get going. And I thought, this is great. Here I am, 32. I'm going to get out of this radio business, which has no future, and I'm going to do a story.

[B]

Well, you were right about that.

[A]

And I. So about three months in, I'm, you know, and beep. You know, this was like a real startup.

[C]

We.

[A]

I remember driving in a truck with. With all the office furniture hanging off the side of it. I'm holding on for dear life.

[C]

Oh, my God.

[D]

We.

[A]

You know, it was like the sawhorse with the doors across the tables and painting the office and everything was a real startup. And. But three months in, it was like, I'm doing shtick. I'm standing up in my cubicle, you know, telling stories and stuff. And I thought, I. I'm a performer. I am not. I cannot do this. I'm a bard. And I quit and went back to radio. Sad. To say the company eventually sold to Macromedia, then became a part of Adobe and is doing quite well. But. And I probably would have been a millionaire by now. But that's okay, because I was about.

[C]

To say it only took four years for it to sell to Macro.

[D]

Find.

[A]

Oh, you looked it up, did you just.

[C]

I just looked it up right now. For five years.

[A]

You're so funny. Anyway, it was a great experience, but it also taught me that you got to follow your heart, not your wallet.

[B]

You want to hear my path not taken?

[A]

Yeah.

[C]

Yeah.

[B]

So I'm finishing my freshman year at Claremont Men's College. It still was.

[C]

I wasn't even allowed to look at a woman for four years.

[B]

Not. No, it wasn't that small.

[A]

Great school. I didn't know you went to Claremont. That's a great school.

[B]

First year. Then I transferred to Northwestern. Foolishly.

[C]

Were you allowed to look at women there?

[B]

Yes.

[A]

That's why Abby got into the woman's scripts. Scripts?

[B]

Pitzer.

[A]

No, not Pitzer scripts, I think.

[B]

Yeah. Yeah.

[A]

And so it's the whole. It's the whole college complex. And I was so excited. I said, oh, God, that's going to be great. Go. And she didn't like it. She went to Bard instead.

[B]

Yeah. Yeah. So summer before I moved, I wanted a job. And there was this internship, political internship. And you go and. And somebody sent me and said, you should go and interview for this. And I go. And you're going to go to all the Democratic press conferences and events and you're going to record things. You're going to record them for creep. The Committee to Reelect the President. No.

[C]

For who? Sorry?

[A]

For the Nixon campaign arm in 19. This would be the 72 campaign, right?

[D]

Yeah.

[B]

I see.

[A]

And. And it was kind of. I don't know why they thought this was a good name. The Committee to Reelect the President. Creep.

[C]

You could have worked on Creep for Dirty Dick.

[A]

Dirty Dick. Dirty Dick's Creep. Did you not take the job?

[B]

No. No, no, no, no.

[A]

Because you're a good. A good Democrat.

[B]

Yes.

[A]

Wow.

[C]

Temporarily. Was this the time where it was coming to reelect Nixon? So, yes, it was.

[A]

I was a kid at the time. I was.

[B]

Oh, yeah, rub that in. Yeah.

[A]

I was campaigning for McGovern at the time. I was going door to door in Daly City, ticky the ticky, tacky houses. Because I wanted McGovern. So, Paris, congratulations on your new mayor.

[C]

Speaking of the one time a week I'm allowed to not wear a burqa. And I think that's beautiful.

[A]

You're not allowed to be political. So we're not going to.

[C]

Yeah. No, that was a joke. Purely based on the strange reaction from a lot of local news organization.

[A]

People are acting like. Yeah, I will say you're going to be a sharia.

[C]

Hottest. The hottest ticket in New York City. The thing you can't find is a copy of today's New York Post cover. Have you guys seen this?

[A]

No.

[B]

Oh, yeah. If you go to my feed, it's in there.

[C]

I would say. I know Jeff has seen is.

[B]

Hold on.

[C]

It's a Post cover that says the Red Apple and it's a photo of Zoron holding up the hammer and sickle. And it says, on your marks, get. So socialist mom. Donnie wins the race for mayor and Zoron supporters are buying it up on masks. It's kind of a. Kind of a banging cover.

[B]

Just put it in the discord. You can show it.

[A]

Oh, my God. It is a banging cover. Holy cow.

[C]

Yeah.

[A]

This is Rupert Murdoch's New York Post. Of course.

[B]

Eat it, Rupert.

[A]

On your Marx get set, Zo. Well, he is. He says, I'm a socialist.

[B]

Democratic socialist.

[C]

He's a democratic socialist. It's different.

[A]

Okay. He doesn't believe that the means of production should be in the hands of the proletariat. Oh, good. Okay.

[D]

Correct.

[A]

I, I think, you know, it's a great experiment with the largest city in.

[B]

The United States, just period. I watched him the day before the election on. On Ari Melbourne and on Morning Joe and he parries anybody's questions. He's just. He's just amazing.

[A]

It's nice to see occasionally somebody who's eloquent and intelligent.

[B]

Yeah.

[A]

Because most politicians seem to be somehow unable to make a simple, simple statement. So that's nice.

[C]

Worth noting that this is, I believe the. I mean, it's got to be. This is the first mayor of New York City that met his wife through hinge.

[A]

Oh, well, there's something.

[C]

A win for all of us out here.

[A]

We don't know who Michael Bloomberg was meeting on Hinge, but anyway, that's good.

[C]

No Michael Bloomberg Field.

[A]

If we're being Bloomberg. Anyway, enough of enough politics. Let us.

[B]

You forgot New Jersey. I got a new government.

[A]

Congratulations.

[B]

The first woman.

[A]

First woman governor in both New Jersey and Virginia.

[B]

We have and showed how. How absolutely flawed polling is and how wrong it is.

[A]

And how did they say that she.

[B]

Was going to lose? We, I, I was nervous. We all thought it was going to be close as hell.

[A]

It wasn't.

[B]

13 points.

[A]

Yeah.

[B]

Ridiculous.

[A]

Yeah. Here is we were talking about Eliezer Yadowski. If anybody. If anybody builds it, everyone dies.

[B]

Drive me crazy.

[A]

Here is a. A reply, a rebuttal from your Cosmotron. Ben Goertzel's substack. He says he's known. Yeah. What does it mean? You're no idea.

[C]

I just think it sounds great to me. I'm sure it means something.

[A]

I'm sure it means something.

[B]

What's. What's the stuff that you get when we hit the singularity?

[A]

Computronium computer?

[B]

Related to Computronium?

[A]

I don't know, but I. That would be a good name for a blog as well. He says he's known Yudkowski since the 1900s. Gertzel started an AI company in 25 years ago, Webmind. They were trying to build AGI in. In 2000 and even then Yudkowski was trying to convince them to stop and. Or think about AGI safety. What's funny is, and this. A number of people pointed this out. At the same time as he says we've got to stop, he also says AGI is the most important thing on the planet.

[B]

Yes. So. Which is because that's. That's this whole. His whole shtick, right, is it's. He so believes in it that if it's not done the way he says, then he endorses thermonuclear war to destroy those who are trying to do it now the wrong way. I mean, you.

[A]

You could say as. As our guest did, that AGI would be the equivalent of the atomic bomb. But that is the atomic bomb. That is actually using the atomic bomb to destroy research is not perhaps the best recommendation for your thesis, Eleazer. So Gertzer says what? Eliezer is right about one thing. We cannot know with certainty that AGI will not lead to human extension. But the leap from uncertainty to everybody dies is a tremendous failure of imagination about both the nature of intelligence and our capacities shape its development. I.

[B]

Don'T know.

[A]

Are you worried? Paris, you seem to be the most worried about this of the three of us.

[C]

About AGI and the impending doom.

[A]

I mean, you're the one who's going to have to put up with it. We're going to be. Jeff and I will be long gone, probably before there's AGI.

[C]

I think there are a million other things to worry about before we.

[A]

I agree.

[C]

I think that's AI's ability to accelerate humanity's already natural tendency to careen towards like global thermonuclear disaster or general, like.

[A]

Exacerbating or bigotry or face recognition or.

[C]

Bad thing here.

[A]

Yeah.

[C]

This technology seems uniquely predisposed to make things worse.

[A]

Did you see the Atlantics piece about Common Crawl? Alex Reisner writing the. We interviewed, of course, Common Crawl, the company quietly funneling paywalled articles to AI developers. It's a, it's a, it's a hit piece on Common Crawl.

[B]

It's very much a hit piece. And so I, I emailed Rich Scranta, the head of Common Crawl, in the morning and he woke up to my message saying, welcome.

[A]

He was our guest a few months ago. Yeah.

[B]

Who put it. Oh, sorry, I just put it in the, in the, in our chat. I'll put it in the next one. So I see who put a stick up the Atlantic's rear end? My question. And he had multiple back and forths, but he said that the article was already written and the view was already set. Journalists do this more than they admit. And he accuses them of lying and other things. So Rich wrote. Rich wanted to kind of ignore it. He wrote a response, which he says.

[A]

He spoke with Rich twice during the reporting on this story.

[B]

Yeah. And the guy kind of ignored it.

[A]

Yeah. So having spoken to Rich, Scrint, I kind of think Common Crawl is a good thing.

[B]

Common Crawl was started because only Google had a crawl.

[A]

Right.

[B]

This was a crawl that enabled others. 10,000 academic papers cite Common Crawl. It's been useful for that for almost 20 years. And then it's open. It's open. Anybody can use it. It's like Wikipedia. Anybody can use it. Right. Wikipedia can make rockipedia. Do you condemn Wikipedia because of that? Because it's open? No. You just recognize what happened. So now the Atlantic's view is, oh, evil AI came along and they didn't stop and they did this, and this is terrible. And they're consorting with the devil, so they must be the devil.

[A]

The devil.

[B]

And then the other complaint here is you have. And Common Crawls put this list out. You have a lot of media companies, starting with the New York Effing Times, who are insistent on being taken down. Now, Common Crawl does not go behind paywalls. It does not sign in as users. It takes things that are free on the web and open. And as Rich says, if you don't want the world to read it, don't put it on the Internet.

[D]

Right.

[B]

And so that's what they capture.

[C]

They put it on the Internet behind a paywall.

[B]

No, no. This is the reason. No, Paris, he does not go behind the paywall. This is not stuff behind the paywall. It's the stuff that is open and visible. That's all they crawl. They do not go behind paywalls. And so the problem becomes then, is that you want, these companies want to take down things post facto. They want to go back and say, take us out of prior crawls, which is difficult. And they're trying to do it, but it's not easy because those crawls are, you know, data sets. But, no, they do not go behind pay walls. Paris.

[C]

So you're saying that the headline from the Atlantic is factually inaccurate.

[B]

Yes.

[A]

Interesting.

[B]

What they say is, what the Atlantic alleges is that if you open an article in certain cases before you get the paywall challenge, you can. The computer is fast enough, it can see some stuff, but it's. This is common crawling behavior.

[C]

If you click the stop loading button before.

[B]

They don't. They don't do those tricks. It's just, this is, this is how crawlers work. This is how Google gets stuff, too. It's how crawlers work. And they just, they're doing the same thing all crawlers are doing. But Atlantic is coming along because it has a dog in this hunt, and it wants to demonize AI and demonize training of AI. This is a hit piece, pure and simple.

[A]

It's a hit piece and it won't be the last. From many outlets. There's definitely a schism now between haters and believers, and there are not many people. I think we, we, the three of us, represent a kind of, this reasonable.

[B]

We represent that debate because Paris and I disagree about this, but it's a healthy debate that we have.

[A]

Yeah.

[B]

The larger question, to me, that's really interesting, and I talked about this on a paper a few weeks ago, where 10 times as many reputable information sites as disinformation sites block all of the crawlers.

[D]

Now.

[B]

That's what's happening. So no longer does common crawl include the New York Times at all. Well, what does that then do? If people are going to use AI and AI gets filled with disinformation, like the web is filled with disinformation, what are we doing to the information ecosystem of society? Do we as journalists have some obligation.

[C]

Are you arguing that all companies then have a moral right to make all of their work available for free?

[B]

No, what I'm arguing is we need a discussion between the AI companies and the content companies, which includes mutual benefit, IT meaning negotiation here and not. And a sane negotiation. And this is why I'm arguing for an API for news where we can go to the AI companies say okay, let's talk, let's talk about payment, let's talk about brand, let's talk about links, let's talk about placement and use and have that discussion so that we can try to include. The other thing I'm arguing, Paris, is that I've seen again and again and again I'm seeing brands, commercial brands, they're dying to be part of AI because they want AI to link to them and they want AI to, to call upon them. And we're seeing responsible news brands cut off AI well that's cutting off their nose, just right their face. And just like the early days of the web and just like the lecht in Germany when news companies tried to say you're stealing our stuff and the Google said fine, then we won't scrape you. And they said well no, please scrape us again because we need you, we need the, we need the eggs old joke. And so I think that what we're going to find at some point is that authors and publishers and news publishers are going to, are going to have second thoughts about their demonization of AI and say, oh crap, everybody's using it and we're not there. And so what do we do? And the strategy they've had in the US is we're going to sue you and we're going to get legislation against you. The strategy in Norway, as I always point out, is no, let's make our own LLMs and let's, let's, let's deal with this at a straightforward level and let's get access to the technology. Let's have a good discussion.

[A]

You know, it's interesting.

[B]

How does society end up better off rather than worse off in the end?

[A]

Darren Okey's positing something I think is pretty important in our discord, which is that news probably isn't important to training a. It's not because news like fish wrap is. News like fish is only a welcome guest the first day or so. There's a lot more other information out there and I think the authors of books, for instance, maybe have more of an argument than, than the New York.

[B]

Times, but they also have more of a need to be called upon. One thing that was found with the Google Google Books project is that when all these books were scanned it increased sales of backlist books because people didn't know about them.

[A]

Right.

[B]

It wasn't there. So.

[A]

Well, one of the things that I think people both love and hate about AI is AI generated video. It's really interesting to See the reaction of people. In fact, sometimes the same person, and I'll. I'll include myself on the one hand, thinks, hey, Sora. And playing with it is really cool and interesting and, And. And at the same time, God, I don't want to see yet another AI Video. So we're going to get a chance to really test this. During the holiday advertising season, Google has created a new AI Ad. Coca Cola's created a new. Here's the Google Ad. Planning a kick, quick getaway. Just ask Google. I won't put the audio on it because I don't want to get taken down, but it doesn't look like it's an AI Ad. It looks like it's a stuffed animal.

[C]

It kind of does look like an AI Ad, though. The motion isn't right for stop motion. It's not right for. For animals either.

[A]

It's too smooth. Yeah. But I think, you know, probably people will go, oh, well, it's just an animated turkey. It's not. I. But on the other hand, the Coca Cola ad. So I didn't. The Google Ad doesn't bother me because it's maybe because it's an animated turkey. The Coca Cola ad I hated. Have you seen this? It is creepy.

[C]

Was watching the World Series, the other final game, and of course there were ads. And I was like, talking somebody. I was like, wow, I never see ads anymore. I'm like, yeah, me either. I'm like, oh, wait, no. The one time I see ads is on my podcast where my two podcast ads show me ads. There's another one.

[A]

Paris, look at this ad.

[B]

Isn't advertising fascinating.

[A]

So Santa arrives as he did last year. Coca Cola did this last year.

[C]

It's so AI it looks terrible.

[A]

It's more AI than a team of.

[C]

100 people that worked on this as well. In addition to the AI if I recall correctly.

[A]

Well, this is one of the stories is Coca Cola said, hey, isn't this amazing? And we used fewer people.

[B]

Go all the way to the end.

[C]

100 is fewer.

[A]

Yeah, yeah.

[B]

Oh, yeah. Coca Cola, of course, is associated with our sense of Santa Claus.

[A]

Yeah, the. The Coca.

[B]

Here's the AI Coca Cola Santa Claus, which I find interesting because it's. It carries it through.

[A]

Yeah. The Santa that we know is. Yeah, there's AI Santa, and he looks just like the one in the ad.

[C]

I think it's notable that despite all of this progress we've seen in AI generated video, these multibillion dollar companies can't use it to make an ad that we all don't kind of agree. It sucks.

[A]

Well, they're good ads. You just don't see enough of them. During the World Series, you're probably seeing the best ads.

[C]

By the way, I did not pay attention to the ads.

[A]

Yeah, those are the best ads. The super bowl, same thing. Those are the best ads of the year because that's the most expensive ad. Real estate of the year.

[B]

But I was talking earlier today with Jason. There's an ad I saw the other day about a mutual fund. And there's a young woman who's at work and then she's talking to a agent and then she's walking down the street and I swear it's AI but.

[A]

How do you know, right?

[B]

You don't know now, I'll bet Paris, you're seeing a lot more AI as than.

[C]

Of course you don't see any ads. The idea of you knowing enough about an ad seemed to describe it is.

[B]

Crazy because we see them 100 times.

[A]

Hey, lady, you're in an ad supported medium now. This is your life.

[C]

I'm not my primary. My salary comes from Consumer Reports accepts no ads.

[A]

Well, I was thinking of.

[C]

This show is great as well. But we are also partially members.

[B]

They're not AI ads.

[A]

No, no, that's true.

[C]

And ads about AI.

[B]

He wanted companies to put their mascots, their advertising mascots up in Sora, which is imagine what's going to happen. So my joke.

[A]

You can, by the way, you can. And I did this. You can now put your animals, your pets, even your stuffed animals into Sora. You can add character soras, which is great. It's so much fun.

[B]

So you will not understand this gag.

[A]

Yes, it's great.

[C]

Well, I don't want to see the weird stuff people would do to Gizmo.

[A]

Well, don't. You don't have to do that. Here. Here I am with my kitty cat.

[D]

Patience, partner. The night is young.

[A]

You're stealthy. This is actually I, I scan Rosie, our kitty cat for this video of me and Rosie hunting mice. It looks just like her.

[C]

Okay, that's a little compelling.

[D]

I see him.

[B]

You like it, Gizmo, you're gonna be a star.

[A]

You're gonna be a star, Gizmo.

[C]

I doubt that they will be able to get her spots right. They never do.

[A]

Well, I, you know, so the way this worked, I took literally a four second video of, of her like running away from me because I was trying to take a video of her and it's, it's pretty good. Her hand. I don't know if Other people can use it, I think is you can. Is Chief Twit Dot Whisker Snow. I don't know why. Whisker W H I S K E R S N O so I don't see any reason not to put my cat's cameo make that public as mine.

[B]

The cat is sure to say, rosie.

[A]

Can I make your public? Lisa's been doing some stuff with her too. She does some cat training videos now, but that looks just like her. I think it's pretty amazing what they can do.

[B]

Don't jump, Rosie. I hate heights.

[A]

That's her look, by the way that she's talking. I don't like her voice. Here, I'll play her voice. I don't like her voice at all.

[B]

Her Sora voice?

[A]

Yeah. Where is it? Oh, I guess it's not coming out. Yeah, it's a fakey voice. But I didn't know how to make her voice anything but. But that I don't know.

[B]

So, Paris, have you ever heard of the Limu Emu?

[A]

Oh, she doesn't know advertising.

[B]

You do.

[A]

You do, right, Paris, When I say Liberty, Liberty, Liberty. What do you think of.

[C]

Yes, I know. Liberty.

[A]

Liberty. Biberty. Liberty. Bibbity Liberty. The emu is. Is also Liberty Mutual.

[B]

So sorry response when asked me the other day.

[A]

She said, what? What would network television do without insurance companies? Half the ads are insurance companies.

[B]

So what I want is a snuff film of the Limu Emu. That's how I want to.

[A]

You're a sickie.

[B]

I'm sick of the.

[A]

But that's what. See, this is why the parent company is not going to put that. What if Scotland wants. Because that's what's gonna happen.

[C]

Kill the. The Limu Emu.

[A]

Yeah, yeah. Lisa tried to make.

[C]

Who do you want to do the killing, Jeff?

[B]

Well, that's a good question.

[A]

Sylvester Stallone? Oh, no. John Wick. Because he's a really good shot.

[C]

I was thinking, you know, the Geico Gecko could get in there.

[A]

Oh, that'd be fun. You're pretending you don't know advertising. No more than you admit to, young lady.

[C]

From when I a was child.

[A]

Advertising is some of our premier creative endeavor in this country. Some of the best stuff is in.

[B]

Imagine the early days of radio where the ad agencies were the ones that made the programs.

[A]

Yeah, what a world. Even the early days of tv. You know, Milton Berle was brought to you by. Who was it? I don't remember, Grandpa. Who was Milton?

[B]

I don't know. Perry. Perry Komo had The craft music.

[C]

I feel like back in those days, the companies that were advertising is like corn made by American farmers. This program is brought to you by Dove Texico Stars.

[A]

Milton Borough was brought to you by Texaco. But there was also, I mean, Philip Morris.

[B]

Oh, yeah. Oh, yeah.

[A]

Cigarettes. Well, I don't, I think Sam is just trying to make more money. Sam has an issue. He, he, he apparently shouted down, no, this must be your story that you put in here, Jeff. When somebody asked him how open AI was can commit to spending $1.4 trillion on training AIs while merely earning billions a year. What did, what did Sam Altman reply to that? Because I can't get the article up. Enough.

[B]

Stop asking me that.

[A]

He literally said, enough, enough. Like enough of you and. Or enough. I'm gonna, we're gonna make enough.

[B]

Yeah. So here's Brad Gerstner on the big two pod was asked the single biggest question that's hanging over the market. How can a company with $13 billion in revenues make 1.4 for trillion in spend commitments? You've heard the criticism. Sam Altman said if you, if you want to sell your shares, I'll find you a buyer. Altman said in response. Enough.

[A]

Oh, boy. That's when a lot of people go on a show like that. See, this is why we don't have Sam Altman on our show.

[B]

Yeah. Because he doesn't handle, he doesn't want.

[A]

Those kinds of questions. Yeah.

[B]

I think there's a lot of people who talk with a lot of breathless concern about our compute stuff or whatever that would be thrilled to buy shares. Altman said. We could sell your shares or anybody else's to some of the people who are making the most noise on Twitter about this. Very quickly.

[A]

Wait a minute. Does Gerstner have shares? Because this is, by the way, not a publicly traded company, so.

[B]

But you can buy all you can buy through the privileged market. We do plan for revenue to grow steeply. Altman said on the podcast. We are taking a full forward bet that it's going to continue to grow. And of course, at the same time, the word is that they're Preparing for an IPO at a valuation of $1 trillion.

[C]

$1 trillion.

[B]

One trillion.

[A]

This. Is this the story from the New York Times how open AI uses complex and circular deals to fuel its multi billion dollar rise. And we've been saying this, that we're.

[C]

Not in an AI bubble.

[A]

Oh, yeah, we're in a bubble. There it is, ladies and gentlemen. Oh my God, her hands are they made of Adamant here. You look good, man. You make a good evil. You make a good bond.

[C]

For anybody listening, I can't describe what just happened because you've got to watch it.

[B]

I'm sorry, can we do it again with a commentary? Do it again, do it again.

[C]

I look down. I hold up a bubble. The bubble's filling up the screen. The words AI Are in it in a bubble. I'm backlit. Suddenly Freddy Krueger s cams come and pop the bubble. I look satisfied at the camera and reveal my Freddy Krueger claws.

[A]

That is amazing.

[C]

That was great.

[A]

You could get a job in the movies with that. I mean, use that.

[E]

Anthony Nielsen.

[B]

Anthony brilliant.

[C]

Anthony bravo. That was incredible.

[A]

But notice how much better than you've gotten even in the six months we've been doing them.

[B]

That's pretty damn.

[A]

That was amazing.

[C]

That was quite good.

[B]

So, so, so, and, and, and let's give points to Bonito. He's been. How long have you had that? How long have you been waiting for her to say that?

[E]

That was actually. That was new today. I was just waiting for.

[C]

I was gonna say Anthony just, just added me in the chat maybe like five minutes ago, saying I should say the word AI bubble.

[A]

The whole thing was a setup.

[C]

And so I did, I did. I'm sorry, I'm sorry to pop your bubble.

[A]

All right, let's take a little break. We'll come back with more. You're watching Intelligent Machines with Paris Martineau, the evil Bond villain with adamantian hands. And Jeff Jarvis, who is a professor emeritus, a villain of his own sort. Villain of his own and the author of some marvelous books, including one coming out this June that is the history of the. Of the Linotype. Linotype, which is more fascinating.

[B]

More fascinating you would think.

[A]

Actually it sounds pretty fascinating if you've seen a Linotype. It's like the most amazing Rube Goldberg invention. It's amazing that it works. And in fact many of its predecessors did not work, including the one in Mark Twain's basement.

[B]

Yep.

[A]

So the fact that it did work was quite an achievement of high tech of another time. Our show to brought to you today by Agency. The Agency building the future of multi agent software with Agency. Agn tcy. Agn tcy. It's now an open source Linux foundation project. Right on. Agency is building the Internet of Agents, a collaboration layer where AI agents can discover, connect and work across any framework. All the pieces engineers need to deploy multi agent systems now belong to everyone who builds on Agency, including robust identity and access management. Very important that ensures every agent is authenticated and trusted before interacting. Agency also provides open standardized tools for agent discovery, seamless protocols for agent to agent communication, and modular components for scalable workflows. Collaborate with developers from Cisco, Dell Technologies, Google Cloud, Oracle, Red Hat and more than 75 other supporting companies to build next gen AI infrastructure together. Agencies dropping code specs and services no strings attached. Visit agency.org to contribute. That's again a G N T C y.org G an open source collective building the Internet of agencies. Agency.org we thank him so much for supporting not only supporting intelligent machines, but supporting a very important effort in in AI. We're going to be covering agents MCP in our AI user group this Friday. If you're a member of the club, let me check the time. Anthony Nielsen and I do this every every month. It's the first Friday of every month. It has become kind of one of the things I really look forward to. It's every month 2pm Pacific, 5pm Eastern, 2200 UTC Nov. 7. If you are, you can watch Live View if you're not in the club, but then it will be on the Twitt plus feed for club members and join us. I think the plan is to get Darren okey. I'm hoping Darren will do this. He just got a new job, which worries me, but I'm hoping Darren will do this.

[B]

I'll be happy for Darren.

[A]

No, I'm very happy for Darren, but he's. He has been writing his own MC Priorities.

[B]

Darren, you know the podcast comes first.

[A]

Yeah, no, the podcast, the unpaid appearance on the Twick Club Twitch shows comes first. Darren has been writing his own mcps. He's been. We've been talking about mcps. Apparently it's not that hard to create one. We'll talk about what an MCP is, how to create it, how to use it and that kind of thing. He has an obsidian MCP that I'm dying to use actually.

[C]

Really Smart guy Club Twit Podcast Mark your calendars. Our upcoming part two of our DND adventure on Monday, November 17th.

[A]

I'm very excited.

[C]

Also to 8pm Eastern, whatever time, 2 to 5 Pacific.

[A]

There I am. The Bard SAG bot on the cheerful playing. My bad.

[C]

I'm going to get my own AI generated image and actually wear a costume this time.

[A]

It looks like they left you out of the prom. The promo copy.

[C]

Wow.

[A]

I might have to add your name. Oh no, there it is. Paul Thorot, Paris Martino, Jonathan Bennett and Jacob Ward. And I will. We're in A corn maze. Micah Sargent by hand.

[C]

Propaganda.

[A]

What is the name of yours? Catheter. Sag bottom. What is no catheter? Catherine Long swallow or something like that.

[B]

Yeah.

[A]

Cathera. How is that pronounced? Kathera.

[C]

Cat. I was saying Catherine, but could be Kathera.

[B]

Did you make it up or did the computer make it up?

[C]

A random generator on D and D.

[A]

Beyond Kathira Long swallow. Right.

[C]

Yeah. Yeah. It's no Sag bottom. The cheerful. That's for sure.

[A]

Well, I didn't. I made that one up all by myself. As you probably could tell, Paul, throughout his Helm Hammer bl.

[C]

All of your great bard quips that you made up all by yourself.

[A]

People didn't like that. So I'm going to have to. I'm just not quick enough on my feet. So one of.

[C]

Right. Like just over the next two weeks, just write single sentence. Five, like two to five quips.

[A]

So one of my spells in this Dungeons and Dragons adventure is something I viciously mockery.

[C]

Yes.

[A]

And the problem was I had to apply it. Vicious mockery to a plow.

[C]

I mean, and you could.

[B]

So this sounds like so much fun.

[C]

The spell is that you mock it so viciously, it loses health. And so Micah was like, leo, like, do you have anything you want to say to the plow? And you could just have said, like, oh, you good. No, good. Good for nothing. Terrible plow.

[A]

I wanted to be clever.

[C]

Instead, he recited a full Chat GPT poem.

[A]

It was really good, I thought. I thought, you know, good.

[C]

But it was. It was too long, so that it was obvious that it was written by something else.

[A]

The work that I did was preparing that and typing in the prompt quickly enough so that I could use it.

[C]

So that is true.

[A]

Yes. See if I can find it. I guess I, you know, I must use perplexity. I don't see it in my Chat GPT. Anyway, it will be a lot of fun. There are part two of the Horror in the Cornfield with Micah Sargent, who has a great scarecrow costume. Mundane of every Once again. Yeah, yeah. He's gonna have to put the makeup on and everything. He had a vampire costume for Halloween. Oh, how did your log lady go?

[C]

Oh, yeah, it went great. So good. Yeah. Let me find a photo. There's one on my Twitter, but there's a better photo on my Instagram, to be honest.

[A]

All right, I'm gonna. Everybody's mocking the fact that I let Chat GPT write my invective. So I will. I will write some on my own.

[C]

You can just be Mocked. That's okay.

[A]

Yeah. I thought it was pretty clever of me to come up with it. I will. I'm not as great at improv as I wish. This is where my dissatisfaction with my own performance on the show. Hey, you really look like a log lady.

[D]

Holy.

[C]

Yeah, that. Someone else found it for me.

[A]

Where'd you get my log?

[C]

Hold on.

[A]

That log is fantastic. You made that? Paris was complaining because she'd been searching all over New York City for a log. You think you could find a log in that big.

[C]

You can't find a log. I mean, now it's kind of. It's been dropped on the ground a lot. So you can kind of see.

[A]

Is it Styrofoam?

[C]

No, it looks Styrofoam that I had. These are two different pieces of Styrofoam that I then glued together and clamped together. Then I carved it from that from a square into round Martha Stewart. And then I added this.

[B]

I think you need a Crafts with Paris show.

[A]

Are you gonna bring that to Florida for Thanksgiving?

[C]

Yeah, I'm actually leaving Gizmo behind, but I'm gonna just bring this log.

[A]

Let's bring the log instead.

[B]

Sure. Did all of your friends get it? The costume?

[A]

None of them.

[C]

It's actually, yes, almost all of them did. But what I'm trying to say is, in. So I live in a specific neighborhood in Brooklyn that is really, really popular on Halloween. Like, it is deep. Throngs of children everywhere. You see, like, difficult to navigate. And I was going one neighborhood over to a neighborhood called Fort Greene, and as I was walking, like, through our neighborhood, like, some people were kind of looking at my customer. As soon as I cross over to the Fort Greene border, they're older. Every single person is like, log lady.

[B]

Wow.

[C]

Great. Log lady. And I was like, wow, I'm among my people here.

[A]

So if you want to be a log lady, go to Fort Greene. And if you're. If you're David lynch, you should be living in Fort Green, obviously. Obviously. Perplexity keeps doing interesting things. They've now got a tool designed to help you find patents. Now you can go to the USPTO website and search for trademarks and patents. But it isn't the best keyword search, right? Yeah, it's not great. So they've launched an AI tool, a patent research agent that lets you use natural language. Are there any patents on AI learning, something like that? Or key quantum computing patents?

[B]

How about the line of type? Leo, can you search for the line of type?

[A]

Oh, should I? I have A Perplexity Pro account. Let me see. Do I have to go to a special line?

[C]

Will this be useful for patent trolls? Probably right?

[A]

Yeah. But it'll also be useful for you as a, as a researcher.

[C]

When am I ever looking up patents?

[B]

When you invent the log maker.

[C]

Yeah, they don't have anymore. You know, they just can't find a blade in your, in your business.

[A]

I think that it's going to come up.

[B]

Yeah.

[A]

Yeah.

[C]

I mean, I feel like the patent. I've searched for patents before.

[B]

I mean, way back when when I, when I was searching on Mergenthaler, the inventor of the line of type, he didn't appear in newspapers at all except in listings of patents in regular newspapers. Patents were news back then. There weren't that many of them.

[A]

What years should. What year range?

[B]

18 to 1880. Not to 1900.

[A]

1880 to 1900. So I'm gonna say show me patents for the line of type or other printing tools.

[B]

Other. Other line casters. How's that? That'll be more specific. See how it does.

[A]

Yeah. Oh, look at this. There's quite a few. Yeah. Here are notable U.S. patents covering the line of type of printing tools focusing on hot metal line casting and contemporaneous printing press tool innovation. See, wouldn't it be useful for.

[B]

Yeah.

[A]

And presumably now I know Paris, you could say, well, how many of these are hallucinations? But it does have links to the actual USPTO entry for that. This is, this is. So I didn't have to do anything special. I just used my regular Perplexity and it automatically.

[B]

They do cool things. Yes.

[A]

Yeah, I think that's really. I think this is a great example of how tools like this are great for research.

[B]

More of this, please.

[A]

More of this? Yes. Well, less, Less Coca Cola ads. Fewer Coca Cola ads.

[C]

Perplexity, though. This week Amazon sued to stop Perplexity.

[B]

Do they sue or do they send a cease and desist?

[C]

Amazon filed cease and desist. This is from Bloomberg. I don't know if it's in the rundown. I'll put it.

[A]

Yeah, at first this is a cease and desist letter, but I guess they've, they've raised a lawsuit and this is awful. So this is.

[B]

Which is awful? Amazon or.

[A]

Yes, Amazon. It's totally anti competitive. Perplexity has an agentic browser comment. I don't use it. In fact, if you're interested, you should listen to Security now from yesterday where Steve kind of says how dangerous these, these agentic browsers can be because they're they're very vulnerable to prompt injection and other hacks. Nevertheless, people can use Perplexity's Comet browser to shop on Amazon. You could say, buy me something on Amazon and it will do that. Amazon, which has its own Rufus, which is an AI agent that does buying and has of course the echo, which does buying, basically says, we don't think Perplexity should be allowed to do this.

[C]

They're specifically accusing Perplexity of committing computer fraud by failing to disclose when Comet is shopping on a real person's behalf, which this is in violation of Amazon's terms of service.

[A]

Well, I mean, and maybe they're going to win on that grounds. Like, just like Apple says, you know, you can't put another messenger app on the iPhone or whatever.

[B]

Here's the question you dealt with this entire career, Leo, on terms of service, is it always found that I am bound by the terms of service of the site that I go to, the company I go to?

[A]

No, no, this is the so called shrink wrap license, which is, you know, just by virtue of using something you, you have agreed to the terms of service.

[C]

I would also say, I mean, notably Amazon in November last year asked Perplexity to stop deploying AI agents capable of purchasing products on the site until the two companies came to an agreement in the practice, says Bloomberg. Like the startup originally complied, but then this August it started using its new Comet browser which had logged into users Amazon's accounts like to do this, which was in violation of their previous agreement.

[A]

But I have to say, it's not like somebody stealing from Amazon. They're using a browser to go buy something from.

[C]

This is kind of interesting context. So like I just said, in August, Perplexity kind of launched this comment browser. This time, Perplexity identified the agents as Google Chrome browser users to yes, it doesn't say it's then when, but Amazon asked them to stop. When Perplexity refused to stop the bots, Amazon tried to block them. Then Perplexity released a new version of Comment to get around the security measures that Amazon introduced.

[A]

So let me just say this. No browser is completely honest in its user agent. If you check what your browser's user agent is, it often says, hold on a second, my mother's calling. I'm not going to answer, but does.

[C]

She want to join the show?

[A]

That might be risky.

[C]

Yeah, that's fair.

[A]

She has moments of lucidness, but then she has moments where she's less lucid. And I don't know which of those moments do you need to pick her up.

[C]

Yeah, we can Van.

[A]

No, I don't need to pick it up. No, no, no, no, no. She, she doesn't, we don't need to go into it. I will call her right after the show. So if you go to what is my browser you can see what user agent your browser passes. So let's, let's parse my own. Now I am on something called Zen Browser but identifies as Mozilla 5.0 gecko Firefox. Yeah, it's based on Firefox but it's not exact. I'm not using Firefox, I'm using a different browser. You could do this with the most browsers, the user agent. There's no law that says the user agent has to say exactly what you are. In fact most of the time it's not. If you Safari it's going to say it's Mozilla.

[B]

Why should Amazon care?

[A]

It's a, it's a.

[B]

Or is it just a way to.

[A]

It's. It's a straw man. It's not.

[B]

It's not.

[C]

One of the things that Bloomberg points out is that shopping agents like these comments comet could one day pose a significant threat to Amazon's really lucrative advertising business. Yes, that's the issue how it makes sense a lot of money by selling prominent placements on its web store in response to shoppers product search queries. If you have a bot shopping for customers then the advertising placement is less valuable.

[B]

Amazon's ad business is larger than the ad business of the higher worldwide magazine industry.

[A]

So they make more money showing you ads when you go to buy something on Amazon highly targeted focused than they do by selling you the product probably. Plus it's more likely that you'll see.

[E]

The Rick like on the store. They still don't make money on the store.

[D]

Right.

[E]

The store is still not profitable.

[B]

Right.

[E]

Like Amazon all makes all their money on aws.

[C]

Well aws, but Amazon advertising has been a big boom for the company especially since the pandemic. Like they make a considerable amount of money doing that. It's not AWS level money, but it's not.

[A]

This is the same thing though. I mean if I use perplexity to do research, I'm not seeing the ads on the websites it's pulling from. This is this, this is the. The whole issue with these AI agents in general is they are disintermediating the advertising on your website. Does the. I guess you're not paying for the content because you're not seeing the ad. But people run ad blockers More than ha. More than half of America blockers now.

[B]

Can you recognize the ads on Amazon? The ads on Amazon don't look like normal ads.

[C]

No, they don't. I believe the markup one time had a really good, I guess investigation or study of this about just the insane percentage of basically everything you're seeing on Amazon's front page is advertising in some way requires the sellers to have paid the company in some.

[A]

Yeah. Well, so for instance, I just searched for spoons on Amazon and there's going to be the overall pick. Well, how does this particular spoon become the overall pick? Amazon.

[B]

It is the essence of a spoon. It's a perfect spoon improve on that spoon.

[A]

And is the Amazon Basics disposable clear plastic spoon the best spoon? I. I don't know. But this is. These are essentially, as you say, and as the markup said, these are essentially paid ad deals. Even though doesn't say advertising.

[B]

If you're wondering what to get Paris for Christmas, I think she needs the spoons because she has these, these sad tin silverware that she still uses.

[A]

What is your, what is your flatware situation, Paris?

[C]

It's a bunch of flatware that has come from other roommates.

[B]

Yeah, she showed us the.

[A]

When you're young, it's only appropriate that you should have mismatched flatware.

[C]

Well, I was planning on bringing for some really fancy Sabra flatware sets, but. But now I can't buy them.

[A]

Oh, they're not in the U.S. yes.

[C]

Paris isn't sending them to the U.S. anymore. And if I do buy them, they're like, they've gone from extraordinarily expensive to like. I can't reasonably spend a thousand plus dollars on flatware.

[A]

You need to have a. You need to have things you need so that people can get you wedding gifts.

[C]

Yeah, because I'm going to be getting married.

[A]

You don't know. You don't know. And I think you don't know.

[C]

You could walk out the door.

[A]

Don't know. Mr. Wright could be standing on your landing right now. No, I'm just saying. Do you have a hope chest?

[C]

What is that?

[A]

Never mind. Kids turn podcast comments into secret chat rooms. This is from Mike Masnick on Tech Dirt. Did I don't know. Have you read in any of our reviews. Have you seen kids exchanging messages?

[C]

I don't think there are any children watching this show.

[B]

Yeah, Paris is as young as.

[C]

I'm probably the youngest person that listens to this show, and that's just because I am on it.

[A]

Well, don't Be so sure. Here's an example. This is an episode of a show called what Leadership Looks like, which I don't think sounds like a kid show, but clearly in the comments there are kids talking to one another. You're gorgeous, Carmen. You're really pretty. Oh gee, she pretty. What's she talking about?

[C]

I mean, this is the YouTube comments. Yeah, sort of thing.

[B]

Years ago when I ran local sports news stories and forums at the news sites nj.com places like that, people would take them over in that case for racist, horrible things before 4chan existed.

[A]

But here's the point. Why do you think kids are doing this? Because they are being blocked from using social media.

[D]

Yep.

[B]

And this is the real they're smarter than the grownups.

[A]

In just a few weeks, kids 16 and under will not be allowed to use social media of any kind in Australia. They will find a way happening. Oh yeah.

[B]

Oh yeah. It's a lot.

[A]

They will find a way kids. And maybe they'll do it in a.

[B]

Hidden way you can't find because they're smarter than their parents.

[A]

Here's the list of social media platforms that will be banned for kids under the age of 16 starting December 10th. Tick tock, Instagram, Snapchat, YouTube, YouTube, Facebook X, Reddit kick threads. This is, this is not. Not. They have to age check. Not that they have to age verify. They're not allowed to use it.

[E]

YouTube Choice.

[B]

They're ready to get ready.

[D]

Yeah.

[A]

Isn't that interesting that kids can't. Under 16 can't use YouTube.

[B]

The other thing is, you know where this comes from. It's not a bubble, it's a moral panic.

[C]

You're sleeping, Bonito.

[B]

I warned you.

[E]

My buttons aren't working.

[A]

Oh, no.

[B]

So title start working.

[A]

But here's the weird thing. Here's what they will be allowed to use. Facebook Messenger, WhatsApp, YouTube, Kids D, Discord, GitHub, GitHub, Lego Play, Roblox, which is problematic, to say the least. Steam and Steam Chat and Google Classroom. Now, of course, when I say it's. It's not age, you know, it's not age verification. It is. What it means is that every single person in Australia will have to verify their age and that they are over 16, 16 or older before they can use YouTube, which is a separate problem, a big problem of age verification. Yeah. December 10th. Unbelievable. Unbelievable. They're already. They're, they're. They're putting up billboards and informational commercials for parents on how to help your teenager get through social media withdrawal. Can you imagine being 15 and said, you know, sorry, you can't, you can't talk.

[B]

So what happens if you're a parent? You say Paris, you can use social media.

[A]

Well, that's an interesting loophole because your parent could say, well I'm going to verify it for you.

[B]

Yeah.

[A]

And we'll just say you're a 32 year old guy from Nazi. And yeah, I don't know. We're gonna have more of intelligent machines. Just a little bit. Pair of smart. No Consumer Reports. What is your new food poison story about? Anything?

[C]

I'm working on it. It's not going to be about poison and food, but it will be about food safety though. Food safety issues. Yeah. And regulatory.

[B]

So did you have. A friend of mine sent me who thinks I'm ridiculous using cacio e Pepe afterwards sent me a story that there's listeria poisoning. Like 18 people died.

[D]

Dude.

[C]

There's so much listeria going on. I'm, I'm subscribed to the FDA alerts. Like it's like I get multiple a day stay away from pre made fettuccine Alfredo products because that's where a lot of them seem to be.

[A]

So listeria is, is a bacteria.

[D]

Yeah.

[A]

And it is like salmonella or E. Coli. It is a contaminant in food. Where does it come from though? How do, how do we. It's just come out of the air. Do we know?

[C]

That's a great question. I don't know. Off the top of my head. It something you want to.

[B]

Where does listeria come from?

[A]

Yeah, it's often in vegetables.

[B]

Where does this listeria.

[A]

And probably the reason for vegetables is they're not processed and not cooked. They don't, they're not pasteurized. They're not in any way, you know, cleaned. I clean all my vegetables in baking.

[B]

Soda, soil and water.

[A]

Can you check animal waste?

[C]

Syria is easily killed by heating foods to high enough temperatures.

[A]

Pasteurization. Yeah. Yeah. Okay. Well be careful. It's out there. Be careful out there. As these to say on Hill Street Blues. Now that we.

[B]

Is your next commercial a food commercial?

[C]

No.

[A]

I, I feel like advertisers should have a, a separation.

[B]

Yes.

[A]

From things that we talk about in the show. Yeah. It needs to be like we need to talk about something nice for a little bit and then we'll do the ad.

[B]

All right, let's find you something nice. Nice and quick. We got, we got it. Google Maps will soon tell you when to switch lanes. Good.

[C]

What?

[B]

Yes.

[C]

That was the Google change law. If you're A new listener to intelligent machines. You're. That's for the fans. If you've been wondering, sand heads out.

[A]

There once, once you're done with your log. If you've been wondering how you should recycle it, maybe you'd be interested in this story. A neural network has discovered an enzyme that can break down polyurethane. Given a dozen hours, the enzyme could turn a foam pad into reusable chemicals. It's a new way to recycle plastics. But what's interesting from our point of view is this was discovered by AI. The tool started working with, the team started working with is called Pythia Pocket. It's a neural network that specializes in a very narrow area. Whether a given amino acid in a protein is likely to contact whatever chemicals that structure combined along with any other functional features. I don't understand what that means. So Pythia, which is a plain old neural network combined with pocket, and then it predicts whether any given protein is likely to form a stable structure. In any event, they used it. The researchers reasoned that a good candidate for breaking down polyurethane would have a number of features. It would look structurally like an enzyme they'd already been working with. But an enzyme that didn't get the job done would also face a trade off between having a structure that was ordered enough to form a similar binding pocket that could have enzymatic activity, but not so rigid that it couldn't fit around the log. Different types of polyurethanes. Anyway. It's pretty technical, but the results were spectacular. Of the 24, the AI came up with 24 highly rated proteins. 21 of them showed some catalytic activity. Eight did better than the best enzyme we've known about up to now. And the best of the designs had 30 times the activity of that enzyme. This is done by AI, so there's some value. I mean, that's not AGI. This is the point, I guess.

[B]

Well, that's, that's why I think these things that are specific are more promising.

[A]

Yeah, very promising, very useful, very interesting.

[B]

And you can trade. It's like, it's like you can train them on an author. You can train them on a task. Yeah, I think that's. That makes more sense to me.

[A]

Let's take a little break. We'll come back with more. You're watching Intelligent Machines with Paris Spoons Martineau.

[C]

I've got a very important announcement when we come back.

[A]

Oh, this is just in breaking news. And Jeff, Hot type Gutenberg Jarvis.

[C]

Jeff holds up stubby hands that are just little tiny keyboard clacks.

[A]

Yeah, yeah, yeah. Little lead type hands. Our show today brought to you by Spaceship. Actually, this is a cool product. If you've been listening for a while. We've been talking about Spaceship and there's a good reason for that. If you need a domain, this is the place to go for domains below market price domains, but also just the best. Darn website is a very modern website and maybe that's why they're doing so well. Spaceship, which is relatively new, just passed a major milestone. Over 5 million domains under management. You don't get that kind of growth by chance. You get it by being the best. Spaceship delivers real quality and features that make sense. And by the way, it's more than just registering domains, but for everything that helps you build and run your online presence. That means hosting, it means business email. It even means tools for creating and managing web apps all in one straightforward platform. Another reason people are switching, of course, the pricing. I always mention the pricing. There is essentially Black Friday and Cyber Monday level values all year round. So you don't have to wait for a sale to get a great deal. But maybe right now you'll be glad to know Twit listeners get exclusive offers that make it even better.

[B]

Uh huh.

[A]

Spaceship.comTwit so whether you're planning a new online project or moving an existing one, Spaceship has what you need to get it launched, connected and running smoothly. A lot more affordably too. I love all of the little tools that Spaceship ads. I love how their DNS works. I love Elf, their AI that helps you do these little DNS chores that everybody hates. Alf loves. Check it out. Spaceship.comTwit to see our exclusive offers and find out why millions have already made the move. Spaceship.comTwit we love these guys. It's my new registrar for everything. Spaceship.com TWIT tech companies.

[B]

Paris. Paris.

[A]

Wait, wait, wait, wait a minute. Break free. This just in.

[C]

I'm wondering if you guys have got me anything because it's my two. It was my two year anniversary of joining the show last.

[A]

Oh, Happy birthday Paris.

[C]

Happy birthday.

[A]

Go out and look on your landing and see who's standing there. It's Zoran Mandani. He's. He's waiting to take you on a date.

[C]

That's very well. And now the co chair of his transition team is Lena Karl Khan.

[A]

What?

[B]

Yes.

[D]

Yeah.

[B]

Yes.

[A]

Oh, that makes me very happy. Lena Khan who was Biden's chair of the FTC and did so many great things which have mostly been undermined by the succeeding administration, including the famous click to cancel rule, which I thought was what every consumer loves, which is the idea that companies should make it as easy to cancel a subscription as it was to create the subscription. You know, you go to a website, you create a subscription and then they say, oh, but you have to call us if you want to cancel it. Or they hide the cancel box somewhere. The FTC made a rule saying, no, no, you can't do that. And of course now history, because the new FTC has said, no, no, no, no, you don't have to do that. Just like the FCC is abandoning the broadband nutrition labels that told you exactly what you were getting from a broadband provider. They don't want you to compare. Anyway, Lena Khan, good news. I hope she can do something for New York. You're, you know, you're living in the greatest city in the world. If you can make it there, you can make it.

[C]

I'll be accepting all cotton based gifts as it is the second anniversary.

[A]

Cotton. You know, the really, the anniversary gifts are not all that hot for many, many years until you get to 50. You know, like wood. There's a clock in.

[B]

She would appreciate wood.

[C]

That is so specific because the other ones are just materials. And then one year is just.

[A]

Yeah, one is just clocks.

[B]

They ran out of things. Corn. Otherwise it would have been corn and.

[C]

Could have been corn. Can be a lot of things. I guess there could be a lot.

[A]

Of food products here. So there's two lists. There's a traditional and a modern list. This is traditional, which is paper, cotton, leather, fruit and flowers. Wood, candy, copper and wool. Bronze for the eighth, pottery for the ninth, tin for the tenth. If you give somebody tin for your tenth anniversary. No. Steel, silk, lace, ivory, crystal. And then it ends at 15 because apparently in the good old days nobody made it past 15.

[C]

Oh, wait, no. 16, Ivy, 17 flowers or fern or I guess this is.

[A]

What are you looking at? There's modern, there's alternate. There's flowers, there's stones. The modern anniversary gifts are plastic, cotton.

[C]

And cotton for some reason, years 26 to 29. No traditional gift, but it picks back up at 30.

[A]

In this they figure, you know, if you're going to make it, you got to get all the way to.

[B]

Those are the, Those are the seven year rich years.

[A]

The 24th anniversary. Musical. Musical instruments. That's a big one. It's not clocks. I apologize for 31st anniversary. It should be timepieces. You give the gift of transportation on the 32nd anniversary.

[C]

My website says it's a ruby.

[A]

I got a one way ticket to Muncie for you. Okay buddy?

[C]

I'm also just trying to the.

[D]

Okay.

[C]

It's interesting that the UK's first year anniversary is cotton. The US's paper. But second it's reversed. But also if you're giving. If it's your first wedding anniversary with your new spouse and you give them like a cotton blanket, I think you're done. So I agree.

[A]

I agree. Flowers and candy, always a good idea.

[D]

Yeah.

[A]

And jewelry basically that's tighty whities.

[B]

Yeah.

[A]

It's just standing instructions are just a jewelry, jewelry, jewelry.

[B]

Then you reach a point in the marriage where enough jewelry, you run out of ideas. Stop with the jewelry.

[A]

Okay.

[C]

80Th traditional.

[D]

How.

[C]

How many people are making it to 80th anniversary?

[A]

You gotta be 100 years old or get married when you're satisfied. Yeah, yeah. How do we get back to AI here? Why don't you guys help me. Dictionary.com is named the word of the year. Are you ready for this? 6 7.

[C]

I saw so many people wearing 67 costumes on Halloween. It was all parents.

[A]

Yeah, because the kids are already.

[C]

Because we were. I was. One of my two parties was a bunch of adults drinking out in a stoop, handing out candy. And every time we'd see parents wearing 6 7, we'd all go 6 7, which is the hand gesture you're supposed to do. And the parents lost it every time. They were so excited. And the kids looked mortified. Which I guess was part of it is it was part of a. It was a parental movement to reclaim 67. So it was no longer funny. And I'm curious to see if it's worked or not.

[D]

Wow.

[E]

Well, that's the secret weapon. That's how you get rid of that stuff is for adults to get into it.

[A]

Yeah, yeah. You start saying skibidy all the time and kids stop saying it.

[E]

I mean Dictionary.com having it as the word that, that basically that makes it.

[A]

It's over. Right. It comes from a rap song. Right? Paris.

[C]

No.

[B]

She'S a little older than the middle school.

[A]

It comes from a skrilla song called Doot Doot 67 where he repeatedly says 67 in his lyrics. Searches for 67 began to spike in June and have been on the rise ever since, increasing six or seven fold. In being named the word of the year, it beat out some tough contenders. This is from Dictionary.com Kiss Cam Tariff.

[C]

Kiss cam. Is that a cold play reference?

[A]

Kiss cam? Maybe. Yeah.

[C]

I mean absolutely. Right.

[A]

Yeah, that's.

[C]

That's so specific.

[A]

A trad wife.

[C]

Trad wife would have been a good one.

[B]

Yeah.

[A]

And the dynamite emoji, which apparently I did not know this has come to signify Taylor Swift and Travis Kelsey. Because it's t. Oh, what happened to would don't we give them a name like, I don't know, Trailor. Travis?

[B]

There was something.

[A]

Swelsey.

[C]

I guess Swelsie could be it.

[A]

Names don't really mesh well. Swelsey. We're gonna call him Swelsea from now on.

[B]

So I just finished reading Stefan Fatsis's Unabridged in which he worked at. No, I don't want to start it again. He worked at Webster. Miriam Webster.

[A]

Fun.

[B]

And tried to get words into the dictionary.

[A]

That sounds like fun.

[B]

And the history of Webster and all that. So it's an okay book.

[A]

One of my favorite books was the book about the oed. What was it called? Had a great.

[B]

Oh yeah, yeah.

[A]

I really enjoyed it. Tech companies don't care that students use their AI agents to cheat.

[C]

Why would they?

[A]

No, it just means you've got skills.

[C]

I mean there's always targeted advertising, Target.

[A]

Open AI and whatnot. During a giveaway of chat, GPT plus to college students said quote, here to help you through finals.

[C]

Yeah.

[A]

Students get free year long access to Google and Perplexity's AI products. In fact, Perplexity even pays refers to $20 for each US student that they get to download the browser comment. The one that Amazon's upset about.

[B]

Corrupt your fellow students.

[A]

Yeah, yeah.

[C]

Gotta be weird to be a student now.

[B]

Oh yeah.

[A]

In a Facebook ad in early October, Perplexity ad showed a student. The Verge has put this in quotes discussing how his peers use Comet's AI agent to do their multiple choice homework. And another ad posted the same day to the Instagram page on Perplexity, an actor tells students that the browser can take quizzes on their behalf. She says, but I'm not the one telling you this.

[C]

Rough. I mean, what are your teachers supposed to do?

[A]

Yeah, I, you know, I think just let you do it. You know the. So as you know, I. Every year I look forward to a coding competition, the advent of code. And one of the things that's kind of marred it lately is AI. You know, people are. Because there's a leaderboard and he has now an faq and he says, the FAQ says should I use AI to solve advent of code puzzles? No. Eric Wastel says if you send a friend to the gym on your behalf, would you expect to get stronger. I think that's a good. That's an excellent analogy. You're not going to get smarter students. Yeah. If you. If you use AI, you can easily use AI to solve these problems, but you shouldn't, because you're not going to get any better.

[B]

But you're gonna win.

[A]

Yeah.

[B]

That's all that matters. Leo, aren't you an American? Don't you understand that?

[A]

Hey, here's some good news. I know many of us have been bemoaning the lack of flying cars, like we were supposed to get flying cars. Joe Rogan's interviewing Elon Musk and Elon says, yeah, we're gonna have. We're gonna unveil a flying Tesla by the end of the year because he's.

[B]

So reliable with all of that today.

[C]

In November.

[A]

By the way, Tesla announced a new Roadster in 2017. They said it'll be ready by 2020. They still have not delivered it. In fact, Sam Altman and Elon got in a little tussle because Sam says, I want my. My deposit back for my Roadster. To which Elon said, you already got it back. Don't give me that.

[C]

So the boys are fighting.

[A]

The boys are fighting. Musk has been talking. According to Engadget, Musk has been talking about flying cars since 2014. But this is the year, everybody.

[B]

It'll fly all the way to Mars, children.

[A]

All the way to Mars. I did have to laugh at this piece from the Baffler, written by Bruce Brace Belden. The hatred of podcasting.

[C]

I loved this piece. There were so many good lines. Hold on, I've got some of my favorite in there.

[A]

So he says, I think I know.

[C]

Which one you're talking about, too.

[A]

The editors at the Baffler want me to talk about my job. They want me to humiliate myself in the pages of this magazine. Very well. I am a podcaster, he says. What is a podcaster? Someone who makes money from talking, often by means of selling dick pills. I don't do that part, but I still obscure what I do whenever possible. Now, I have to point out this guy thinks podcasting began three or four years ago with cereal, and he thinks it's all over now because I guess he doesn't.

[C]

I mean, he. He describes how one of the first podcasts he listened to is in, like, 2008.

[A]

Okay, okay. But podcasting during. Obama had a wondrous feeling. He writes Americans. The shows that were popular around then, like Cereal or Invisibilia, which debuted the following year. The pledge to explore the intangible forces that change shape human behavior. All at a very Eagle Scout approach. I heard it on a podcast. People would say in 2015 that if you said I heard it on a podcast, you were trying to sound smart. In 2025, it's better to lie.

[C]

That's true. What were your favorite lines? Is then Covid hit and podcasts went through the roof. And now I have the salary of a dermatologist and live in Brooklyn, which is Israel for podcasters.

[A]

It's our home. Our special home. It's true, we do have probably a larger proportion of people from Brooklyn on our shows than any other city. That is probably true. It's very frequent that we will have somebody from Brooklyn on a year from Brooklyn.

[B]

I used to live in Brooklyn.

[A]

Jeff's from New Jersey. But that's, you know, Brooklyn, Brooklyn West.

[B]

I round up.

[A]

Yeah, roundup. That's right.

[D]

I don't know.

[A]

I just thought it was. Thought it was funny.

[C]

And I also, honestly, I'd really recommend this piece. It's a fantastic. It's written by one of the hosts of True Anon, which is a popular podcast. But he kind of weaves in both, like, kind of memoir ish, writing about his really interesting backstory as like a member of the US army who kind of s. Posted his way to popularity, for lack of a better word, then became a podcaster instead of, I guess, going on a different sort of tour of duty and has ended up in this strange life. But it is also just kind of an interesting reflection in the last, like 10, 15 years of podcasting, especially in like the post pandemic era.

[A]

I think my kind of attitude towards podcasting is. Is colored by the fact that I came from radio and it's just radio over the Internet as far as I'm concerned.

[B]

Well, Howard Stern always argued that podcasting isn't radio, and you told us you really work on radio, you don't understand. And he made fun of podcasts for years until now.

[C]

Well, I think that's the thing is part of this, what this is like chronicling is the shift from what used to be considered like these highbrow, like capital P podcasts that were produced, shows that seemed like they were really there to learn to a. Like to this, to this this. He literally says, already isolated individuals now work from home and the only room with only a roommate or dog for company. The sharp rise in loneliness facilitated the shift from highbrow liberal shows to cartoonish hangout sessions with the worst. Every gender has to Offer boy.

[A]

That describes this show in a nutshell. This is a cartoonish hangout section session.

[C]

Yeah.

[A]

But the thing is, this show has always been that. It's.

[B]

Since.

[A]

We'Ve been doing that since 2008. So it's just, you know, trends. There's been. There's been real podcasting all this time, and then there's been the hyped podcasting, the celebrity podcast. They're still around, but, you know. Anyway, TikTok has announced its first awards show in the U.S. by the way, did. Did Trump not meet with President Xi in South Korea last week?

[B]

It didn't come up.

[A]

Tick Tock did not come up.

[B]

No. Neither did Taiwan. Those are the two things that people thought were going to come up.

[A]

Tick Tock and Taiwan.

[B]

Important issues in life.

[A]

Yeah. Because he said, we have a deal. What? It's still in limbo. In other words, there's no. Well, they just. They decided to go ahead. They're going to do a. A Tick Tock award show. It'll have creator of the year, video of the year, Muse of the year, breakthrough artist of the year. This is just weeks after Instagram announced its awards program. There will be an actual award show, though, for TikTok, including a red carpet, live performances in a live audience. December 18, the Hollywood Palladium in Los Angeles. It'll be streamed on TikTok. Oh, and on Tubi. And then on to be. To be. I want to call it Tubby, but it's.

[C]

I want. I want to call it Poob. Have you guys heard of Poob?

[B]

I don't think that's right for our audience.

[A]

Was that. Wasn't that what Clarence Thomas found in his Coke can? What is Poob?

[C]

This is from. I guess maybe it's a tweet or maybe it's a Tumblr post. It's become like a meme where the tweet just says, have you seen the new show? It's on Tubu. It's literally on Heebie. It's on Pootie with ads. It's literally on Dippy. You can probably find it on Oui know, dude. It's on Gumpy. It's a pheno original. It's on Poob. You can watch it on Poob right now. Log on to poop right now. Go to Poop. Dive into poop. You can poop it. It's on Poop. Poop has it for you. And I love that.

[B]

That's beautiful.

[A]

This is on Tumblr, actually.

[C]

It's one of those phrases that lives in my.

[A]

That is. Is it's on Pooh and it really is. What's going on? You know, it is. We had a bit of a crisis in the Laporte household on Monday night when. When it was. We discovered that we were not going to be able to see Monday night football because ESPN, Disney, Hulu have FUBU, which is one of the Disney channels. FUBU, it's on FUBU. They pulled off of YouTube TV and they're not going to. I don't. So we've seen these carriage arguments before, right? This one's different because Disney has fubu, espn, Disney plus Hulu, it has four massive streamers. They would far prefer you watch football on their streams and on YouTube TV. And I wonder if a deal is going to be made, which is going to be a big issue for YouTube TV.

[B]

They begged YouTube Google to put up ABC for election night, and Google said no.

[A]

So who. Who is it you think that is in the driver's seat on this one?

[B]

I think it's two monstrous companies. It's just like the shutdown.

[A]

And as always, when the. When the giants battle, it's us little people that get stepped on.

[B]

So you couldn't watch any football?

[A]

Well, we. So fortunately, ESPN does not have all football. Most of the footballs on Fox and CBS and other networks. But. But Monday Night Football is an ESPN production, so that's the big one.

[C]

Sports game. For the first time potentially in my.

[A]

Life this weekend, the world serious.

[C]

The world serious. Did you enjoy? Was riveting. I had to have someone explain to me how baseball worked throughout it. But once I understood, wow, what.

[A]

By the way, that person had the best night of their life.

[D]

I was.

[C]

But he did actually.

[A]

He was like, well, let me. Me tell you what's going on here. The runner on third is threatening to run to home.

[C]

He was like, paris, they're not called points, they're called rums.

[A]

And I'm like, best night of his life. Finally, somebody I can tell about baseball.

[C]

I realized that my friend is an inherently uninquisitive person. Because I was like, why is it called the count? And he's like, I don't know what you mean. He's like, I've never thought about that before.

[A]

He didn't count.

[C]

And it always no balls.

[A]

One ball, two balls, a strike, two balls. It's the count. It's three and two is the count.

[B]

Fairly obvious.

[A]

You're counting, counting how many balls and strikes.

[C]

Yeah, I know, but the count is just a silly name for it.

[A]

Oh, baseball's Full of them. Oh, the bunt. The suicide squeeze.

[C]

I like that. They had to introduce a clock to stop people from just wasting everyone's time.

[A]

It was so silly.

[C]

I think that's pretty cool.

[B]

But then so boring.

[A]

It is relieving.

[C]

I will say the one game of baseball I've ever truly watched.

[D]

Riveting.

[A]

Well, you watched literally the most exciting game of the year. Did you watch game seven?

[C]

Yes.

[A]

Yeah.

[C]

Literally a bottom of the ninth, bases are loaded situation.

[E]

Possibly of all time.

[B]

The most possibly.

[C]

It was great.

[A]

I had a lovely experience. I will say this. I was rooting for Canada because I am an honorary.

[C]

As was I. It also took me a minute to be like, why are there Canadian teams in baseball? In American baseball? But I guess why.

[A]

Why are half the pitchers on the Los Angeles Dodgers from Japan?

[B]

Yeah.

[A]

When they come to the mound, they have to bring an interpreter.

[E]

Because the Japanese are the best baseball players.

[A]

They're the best. Look at Ohtani. Oh my God. He's a great pitcher and a home run hitter. But that doesn't HAPPEN in the U.S. you're one of the other. You got to pick a lane.

[B]

All right, so I think Google's space data centers. Did you see that?

[A]

Yeah. This is not. They're not the only ones. This is. This is. The next big thing is data centers and data centers in space.

[B]

I found that interesting.

[C]

I'm just trying to imagine a situation where we lose a data center because it just goes like somewhere into space.

[A]

Well, there, but see. Oh no.

[C]

My data center collided with all of the space junk. Oh no, my data center collided with the Tesla road.

[A]

Oh, it's way out there now. But I mean.

[C]

Wait, well, is that thing still tracking where the Tesla is?

[A]

Oh, yeah. Oh, yeah. Let's see. Tesla, it's kind of. I don't know if it's within radio. I think it's in orbit now. Not of the. Of us. It's of the sun. Yeah.

[B]

Where is Starman?

[A]

Starman?

[B]

Where is roadster.com?

[A]

Here we go. Yeah, it's in orbit around.

[B]

It's nearest Venus right now.

[A]

Yeah. Oh, Venus. It's a. It's 1797-771797-77823 miles from Earth. Moving it away as a speed of 9,388 miles an hour.

[C]

That's beautiful.

[A]

It's moving towards Mars. Yeah. It has achieved a fuel economy of 30,811.9 miles per gallon.

[C]

Great.

[A]

If the battery were still working. Remember Starman is listening to David Bowie's Space Oddity over and over and over again. It would have been 768,000 times by now.

[C]

Space Odyssey in one ear and is there life on Mars in the other ear while the stereo plays his Hitchhiker's Guide to the Galaxy.

[A]

Wow.

[E]

But honestly though, all the radiation probably killed all the electronics already by now.

[C]

Yeah, yeah.

[B]

Plus there's no sound in space, so.

[A]

Yeah.

[C]

Your telescope would have to be 42, 000ft in diameter to resolve the upper stage from Earth.

[A]

Seven years, eight months out there. Wow. But. But back to Google.

[B]

Yes.

[A]

You know, I mean, this is the thing. Stuff's. I didn't know this, but apparently Starlink loses one or two satellites a day reentering the Earth's orbit because. Because SpaceX has made it so cheap to launch these things, relatively cheap to launch these things. They just keep. They know they're going to decay and they just put more up, putting up more faster than they are losing them.

[B]

But where does this. Does this orbit Earth or does this orbit the sun?

[A]

No, no. Or see Earth. It has to be close, otherwise, you know, you don't. It's probably low Earth orbit, I would guess. I haven't looked at this yet, but because you, you don't want a lot of latency, it's going to be sending data down.

[B]

Right.

[A]

Five or six years. They expect the lifespan up there because the radiation will.

[B]

Yeah, they're not sure what the radiation will do to the chips.

[A]

Yeah. But Google is not the only one I think is either Japan or China has been planning to do this also.

[B]

It takes away problems of cooling it.

[E]

Well, you'd think so, but it does different problems. Yeah, it creates.

[A]

So you would think so because they're in space where it's very, very, very cold. Right. But because it's a vacuum, heat is not conducted off, it has to be radiated off. So actually it isn't as easy to cool it as it would be on Earth. You have to have, you know, special technologies to cool it. But of course there's a lot of sun. Google calls the project Project Suncatcher. They're going to launch two test satellites in 2027, each carrying four TPU units. And then as space transport comes down in price, they think this will become economically viable in 10 years, 2035.

[B]

They took chips to UC Davis and used a particle accelerator to irradiate the processors and simulate the years of solar exposure in space.

[A]

Right.

[B]

They held up quite well.

[A]

Well, there you go. Well, watch out because AI, according to Anthropic is becoming introspective. This is such bs.

[B]

Wait, you're the sandman. You think that all this is true?

[A]

No, I think this is more anthropomorphization. On Wednesday, Anthro Anthropic published a paper titled Emerge. Last Wednesday, not this Wednesday. Emergent Introspective awareness in large language Models in some experimental conditions, Claude, appears to be capable of reflecting upon its own internal states. See? Or is it just repeating words? Words that are next according to probability. You're ascribing kind of an intent to random words. I. I just don't. Doesn't make sense. It turns out though, the. The paper comes from a computational neuroscientist and the leader of Anthropic's model psychiatry team, Jack Lindsay, wrote, our results demonstrate that modern language models possess at least a limited functional form of introspective awareness. That is, we show that models are in some circumstances capable of accurately answering questions about their own internal states. And probably inaccurately as well, be my guess. I just feel like that's more anthropomorphism. And it's also, of course, Anthropic's interest to gather. They're alive, they're thinking and.

[B]

Wait, what about thinking?

[E]

What is the they there? Is it like the individual chatbot you're talking to? Or is it like all the chatbots is one person? Or like, what. How does that work?

[A]

It's a machine.

[B]

Yeah.

[A]

So it's one machine you're typing at.

[E]

The one machine that I'm on is the. Is the.

[A]

Is feeding back words?

[B]

Yes.

[A]

That sound like it's thinking about itself.

[B]

We caught it thinking. We thought. We caught it contemplating. But for all of that, Anthropic might have a better business than.

[A]

They're making money.

[B]

Open AI. They project $70 billion in revenue, 17 billion in free cash flow by 2028.

[C]

They project, based on what I project, that I'll have $70 billion in revenue.

[A]

This is their most optimistic growth forecast.

[C]

That's also my most optimistic growth forecast.

[B]

Therefore, on corporate companies rather than on public adoration.

[A]

Yeah, there's an interesting conflict here because on the one hand, there's no question that companies want to use AI. They're paying for AI. Individuals are paying for AI. I have 4 or 5 20amonth accounts because that's you. But. But I love this stuff. I'm using it all the time.

[C]

Which one's your favorite right now?

[A]

Depends what I'm doing.

[C]

I really like distinctive favorite.

[A]

Well, for coding, Claude code is easily My favorite. I used to really like Perplexity, and I still think Perplexity might be the best, but they've soured a little bit on the company itself. So I, I use chat GPT5 a lot. I, you know, I have that, that Common Lisp GPT I, I created many moons ago, probably a whole year ago when they had first had custom GPTs, maybe two years ago. And, and I've updated that a little bit to include, you know, more information and to use the latest chat GPT5. And it's, it's really useful for asking questions. So I use it quite a bit, you know, when I do, when I. I'm getting ready for the advent of code December 1st. So a number of us, Paul, also known as Chocolate Milk Mini Sip in the Discord, and I are doing another coding challenge for. That's for the month of November, called Everybody Codes. And it's. The problem is, you know, and I think this is true for anybody who doesn't code. Randall Schwartz, who's a pearl wizard and former host of Floss Weekly, told me that if you've got, if you're going to keep using be a coder, you got to use the language two or three hours a day or you'll lose fluency. Like any language, you know, you got to keep speaking. It's reaching over to get my. So often as I'm working, I will, I will refer to books like this and I'll go, okay, what is that? What is the syntax for that command? But now I, it's much easier for me to just ask Chat GPT. I've got this book and many others fed into it, and I'll just say, hey, I want to. What's the command to do this? I want to do this. How do I do that?

[B]

Jeremy's example, and he's vibe coding.

[A]

Well, I see. I don't want to vibe code it because I want to actually do the coding because it's like going to the.

[B]

But here's a guy. Here's a guy who's doing the real serious stuff.

[A]

I know it's pretty cool using language.

[B]

Just English.

[A]

Yeah.

[B]

It's philosophy majors when Paris and I both graduate with our PhDs in Philosophy.

[A]

And as you know, Anthony Nielsen's been using Claude for research for guests. He generates a wonderful research precis. He did it for our last guest. That it makes it very easy to, you know, understand what they're going to talk about and get questions that I wish I'd had that my entire career would have been Hugely useful. So I use a little bit of everything.

[C]

Join Claude. Lately I've been, yeah. Comparing Claude and Chat GPT, both the.

[A]

Pro versions for images I use.

[C]

I will say just putting in this AI introspection study and asking Claude what it said, because I think I'd seen somebody on Twitter Blue sky post the their screenshot of Claude's thinking on this same introspection. And I think theirs was like, oh, no, I can think, or something like that. Am I going to get a. A funny response like that? And no. I've trained Claude too, that it literally in its thought process, like, do not be cute or cloying in your response.

[B]

That's for boring Claude.

[C]

Yeah, it annoys me when it tries to be cute or use emojis.

[A]

This is the image I had Chat GPT generate. You know, you can ask it, well, what do I look like? And it said, well, give me a headshot. I said, no. And I said, okay, well, here's what you look like. Podcaster hero Leo Laporte. And it comes with accessories.

[C]

Why did you have Dune behind you?

[A]

I don't know. Well, maybe the sand. I don't know.

[E]

And town.

[A]

And townation. Townation yet. Sci fi. I must have asked it. I don't know. I have a cast iron. I have a cast iron pan and a basting brush attached to my coat.

[B]

Always handy to have both.

[A]

Yeah.

[E]

Oh, it's a mashup beach without them. It's a mashup of a cartoon. And the toy. The toy meme, everyone.

[A]

Yeah, the action. Because I had it once do that. Yeah. So basically it says, I can draw you, but to make it you, you, not the guy who vaguely podcasts, I need one clear photo, front facing, good light, no heavy filters. Head and shoulders. Shoulders is perfect. And by the way, because I once asked it for a stipple portrait, it said, I'll. I'll default to the stipple portrait if you don't pick a style. Isn't that it? So it. It. It is remembering everything to make it you. You.

[B]

It speaks like that.

[A]

Yeah, it's pretty. Pretty.

[C]

Not. Guy who vaguely podcasts.

[D]

Yes.

[A]

Yeah, yeah. I. I think they're very. Look, I know how it works. It's very hard, I understand, not to anthropomorphize and say, wow, this is like a real person I'm talking to. I understand that. And that is problematic, for sure. So what are you doing with it, Paris? What do you like to do with it?

[C]

What have I done recently? I'm kind of doing an impromptu road trip this weekend. And so I asked both Claude and Chat GPT to analyze my route to. And I don't know, highlight any places that think I should stop between here and Vermont. Oh, I am. This is unrelated to them, but it was already on my list as I'm going to the American Museum of Tort Law, which I've been wanting to do for years.

[A]

Wow, that'll be exciting.

[C]

It's actually going to be so gripping.

[B]

Paris, Paris, Paris, Paris, Paris on Saturday.

[C]

I want to go to the printing museum outside. It's. I don't have the time for it. I really tried Paris. I know, but I do. I did mention this to my two graphic design a couple and they're all like, we've got to go.

[B]

So I'm gonna get amazing.

[C]

I'm gonna go with them.

[B]

Okay, good.

[C]

Okay. You gotta look at their merch because they've got both T shirts that have flaming Ford Pintos on it as well as one of the flaming rats. And I'm gonna get a lot of it. It's a raider sponsored museum.

[A]

Don't.

[C]

Don't spoil it for me.

[A]

Oh, don't look. Don't look. So tort law is what? Lawsuits, right?

[C]

Yeah, it's like product liability.

[A]

Look, this won't spoil it for you. Here's the umbrellas in the doorway.

[B]

Because you don't want to poke someone in with an umbrella and then to get sued.

[C]

I mean.

[A]

Yeah, it's lawsuits.

[C]

It's. Well, it's law that. Yeah, it's relating to liability. There's stuff like. It's liability law. Like you've got tobacco cases, Ford Pinto stuff with asbestos.

[A]

Actually, this sounds like a great museum.

[C]

Actually gonna be. I legitimately have been wanting to go for years, but it's never Connecticut.

[A]

Winstead, Connecticut.

[C]

It's like two hours.

[A]

Isn't that a great name for Connecticut? I'm.

[B]

You could also stop off at Mark Twain's home, which is in Hartford. Hartford, yeah, just up.

[A]

I will be going through that.

[B]

I mean, listen, it's pretty fun. They also.

[C]

My oyster.

[B]

You will see the. The page machine that bankrupted him in the basement.

[C]

I am. One of the reasons I'm going is I'm going to a famous puppet museum at some point as well as.

[A]

You know what? You're a character, young lady. You are a character. Puppets and torts.

[B]

Tales to tell.

[C]

Puppets. Torts. Gaelic festival of death. You know, we love to see it.

[A]

Puppets, torts and skorts coming up.

[C]

Well, not squ. It's going to be Cold.

[A]

Yeah, no squirts. Ladies and gentlemen, our picks of the week are coming up momentarily. But first, a word from our sponsor. You're watching intelligent machines brought to you by Monarch. Oh, I love Monarch. I use Monarch all the time. Wouldn't it be nice to kind of know to feel confident and organized in your finances with Monarch? Monarch is an all in one personal finance tool that takes your entire financial life and brings it together in one clean interface on your laptop. You can I have it on my phone right now. Just for our listeners, Monarch is offering 50% off your first year. Just use the code iamonarch.com and do start at the website. That's the best way to set up Monarch. It's the easiest. And then you can use it on other devices. But, and that's the nice thing is that your entire financial life is available to you no matter where you are. Monarch is built for people with busy lives. And if because you're busy, you've put off organizing your finances, Monarch is for you. Monarch does the heavy lifting. It's very simple. I link all my accounts. Just took a few minutes. They do it very securely. And then you're going to get clear data, visualizations, you know, graphs. You're going to get smart categorization of your spending. I really like that feature where it knows it does budgeting for me. I don't have to think about it so I know exactly what I've spent on each category, how much I spent eating out each month, how much I spent on, on capes each month. You know, that kind of thing. What's my hat budget for this year? Real control over your money. You'll never need to touch a spreadsheet again. No more entering by hand. I used to do this back in the day. That's what you used in a computer. You would get your checking account statement, you'd enter in each transaction. You don't have to do that anymore. You link the accounts, you're done. And you always know exactly what your net worth is. You know where your money's going. It's not just another finance app. It's a tool not only used by people like me, but real professionals. Financial experts love and use Monarch. Wall Street Journal named it the best budgeting app of 2025. I almost don't want to call it a budgeting app. That's just one of the many things it does. Forbes said it's the best app for couples. You know, they have a really nice feature where you can invite your partner to have access to your stuff and you control what they can see and can't see it. It is really important for couples to have those financial conversations. Often, you know, people avoid that. This is a great solution named in CNBC's top fintech companies in the world. And by the way, you're going to want to follow the Reddit community, a very passionate community, 34,000 users. And it's more than just people talking about Monarch giving you ideas about how to use it. They're there, they shape how the product is developed. The Monarch people, people pay close attention. Money can can really become a problem with couples but Monarch brings them together. It gives your partner full access to your if you wish to your shared dashboard, including linked accounts, budgets, goals, spending activity all in one place. No drama, no, you know, no heaviness, just information. And it's no extra cost for you to add your partner. You can even give access to your financial advisor also at no extra cost. That's really handy. Don't let financial opportunities slip through the cracks. Use the code I [e-mail] in your browser for half off your first year, 50% off your first year. Monarch.com the code is I am highly agree really a great tool. I just opened just now to see what my stock market is doing monarch.com all right, ladies and gentlemen, time for Are you asking Paris where to go for pizza in New Haven?

[C]

I'm ask. Yeah. Where should I go?

[A]

Don't ask. AI man, you got where should I go? Experts on his show where should I.

[C]

Go in New Haven once there's no.

[A]

Question you should be go to Pepe's P E P E. Pepe's tomato pies are the best pies.

[B]

If you're in the mood, get the clam pie. Get the clam pie.

[A]

If you're in the mood, get the clam and garlic pie. One of the great things is you eat it one evening. It stays with you for days. Pepe is was founded in 1924. It's the original pizza place. There's another place you should go if you can spend a little more time in New Haven. It's called Louie's Lunch and it's mentioned in the Whiff and poof song from the tables. Animore is to the place where Louis dwells. It is the birthplace of the hamburger. Established in 1895. It'll be the best. Look at these. These are the grills on the left.

[C]

Okay, that's actually I might do that instead.

[A]

It's better than pizza. Well, no, it's not. Nothing's better than pizza.

[B]

But I think you're gonna get complaints for having just said that, Lena.

[A]

It is an amazing. What's funny is they grill it in these grills that have been there since 1895, gas grills. They're grilled sideways on toast, and they put Velveeta on it, and it's still the best thing.

[C]

I'm excited by this already. I'm excited. I was sold once I saw that their website says birthplace of the hamburger Sandwich.

[A]

Yes.

[C]

Language I need.

[A]

It's not on a bun, Louise. I would.

[B]

Is a hamburger a sandwich?

[A]

Yeah.

[B]

Yeah.

[A]

I mean, it is. The menu is very simple.

[C]

What would a hamburger be if not a sandwich?

[B]

It's a hamburger.

[E]

It's his own category. It's a subcategory.

[C]

Yeah, it's a subcategory.

[A]

Look at this. It's on a piece of white bread. It is a.

[C]

It's like saying a grilled cheese isn't a sandwich.

[E]

Okay, then what is a sandwich? Jeff is a blt, not a sandwich. Because it's a. I don't know, whatever.

[B]

No, if you're between slices of bread, it's a sandwich. If you're in a bun, it's not a sandwich.

[A]

So Louie's is in a bonus bread. This is, by the way, the worst picture. I don't know why it's on their website.

[C]

It is horrible. It's really compelling to me. The fact that the website is terrible is a real veteran. My cat.

[D]

It is.

[A]

It is. Here's the. Here's the menu. You can get your. You get burger, potato salad, homemade pie, Poland spring water, Pepsi, Diet Pepsi, Snapple or Fox on Park Soda. Pepsi, Pepsi. No, no, no. Coke, Pepsi, Pepsi, Chiburger, Chip, Burger, Chip.

[B]

But the meat, she has no idea.

[A]

They actually.

[B]

John Belushi. Is this another ad in Saturday Night Live? No.

[C]

Okay.

[A]

Many years ago.

[B]

Anyways, I used to go to all the time in Chicago across from the Chicago Tribune in the Second City.

[A]

Oh, that was a real place.

[B]

Oh, yeah. Oh, absolutely.

[A]

Yes. Great. Anyway, Louie's lunch and then Pepe's Pizza. If you. Pepe's usually has a line. There's also John's. A lot of people like John's. I think Pepe's is the original. And Jeff, who is a pizza expert, you agree with.

[B]

I say I judged it.

[D]

I didn't.

[B]

It didn't win my contest.

[A]

But who won your contest?

[B]

Gino's. East Chicago.

[A]

Yeah, I like Chicago style. This is not stakot. This is the opposite of it. Napolitana. This Is a Naples style style, traditional style pizza. It's incredible. You. They don't actually put tomato sauce on it unless you order it.

[C]

Interesting.

[A]

Yeah. Oh, but. Oh, my God. Let me see if I can find.

[B]

I've had the.

[C]

You can't be showing this to me at this hour of the evening.

[A]

I know. This is cruel.

[B]

This is really all.

[A]

They don't even. They don't even have it on their menu. Because you have to know. You've got to. To know.

[B]

Oh, really?

[A]

Well, I don't know. Clam and garlic is the one you want.

[C]

Pizza. I was eating egg rolls throughout this entire recording, but now I'm hungry.

[A]

So let me tell you about this pizza. They cook it in a wood fired brick oven. It's wood and it is burned on the bottom. You know how it gets little burnt bubbles on the bottom? It makes it so good. Oh, my.

[C]

You guys are crispy pizza folk or chewy pizza folk.

[A]

This is chewy. It's not crispy, but it's thin. Thin, thin crust.

[C]

Okay.

[A]

Yeah.

[B]

Paris, are you a crispy or shoot?

[C]

I'm a crispy.

[A]

Oh, I'm more sure crispy. No, no, no, no, no, no.

[C]

I think that like crunch podcast. I've said this view in the podcast. I think that everything should have a little bit of crunch in it.

[B]

I mean, then you can't fold it the way a New Yorker does.

[C]

Well, that's the thing is. I guess it's fine. I. Yeah, you do fold it, but it kind of like snaps.

[B]

Yeah. And then everything drips through it. No, it's not.

[A]

Oh, that's because you grew up in Florida. I'm sorry. This is not okay. You. You. We gotta. We gotta fix this.

[C]

I. Most the pizzas I eat and enjoy are not crispy. But I'm gonna say if it's. When. When people say chewy, I think of like a thicker pizza.

[B]

And I don't know, it doesn't have to be.

[A]

No, no, no. This is thin. The crust. The crust is chewy. The crust has a wonderful chew to it and a crunch.

[B]

I want the kind of artsy pizza that I really. The brick oven or. Or pizza.

[C]

I live near a lot of fancy pizza places and I. I eat well.

[A]

No fancy pepes.

[B]

All right, Paris, have you had John's Paris.

[A]

Next door to this famous sandwich shop they call Salt Hanks. Have you ever been there?

[C]

Oh, no. Oh, yeah. Yeah, back in the day. It's fine.

[A]

Yeah, I gotta come. I gotta just fly out for.

[C]

Why are you hang out with us? We should okay. Fly out, do the Amazon week and then get our little studio and then we could all. What if we all did a show in person?

[A]

We did that once, remember, Jeff? We did it at the.

[C]

With me.

[B]

That's right. Yes, we did.

[A]

Yeah, it was with Gina. Right.

[C]

I could bring props. We could do prop comedy.

[A]

That would be fun. That's very tempting. Where can we get a studio? Okay. Anybody?

[C]

What if we did. What if we did a New York City live show?

[A]

Oh, I would love to do that.

[C]

That'd be so fun.

[A]

We get five or six people. It'd be so amazing.

[C]

It'd be great.

[A]

We could. We could sell out Lou's lunch.

[C]

More than Lou's lunch.

[A]

All right, Paris, your pick of the week.

[C]

For once it's reversed between me and Jeff. I've got like four picks the week and Jeff only has one. Leo's immediately gone to the strangest one, which is another thing I've was thinking of stopping on my trip, which is this thing called the dog Chapel at Dog Mountain in Vermont, which is. In 1998, a folk artist died and came back to life five minutes later. He, upon her huggery stated that the near death experience had a profound effect on me as an artist. He realized that he had to build a chapel, one that celebrated the spiritual bond we have with our dogs. And it would be open to dogs and people of any faith or belief. That system. It's apparently a chapel just filled with notes from people about their dogs.

[A]

I like to sign up front that says, welcome all creeds, all breeds, no dogmas allowed.

[C]

Kind of cute, right?

[A]

Yeah. I think this sounds. It's very sweet when you get to these places though, because you've done this before. Do you get there and go, okay, I'm here now.

[C]

I get there. I go around. I'm probably going to spend a good 15, 20 minutes reading all the different notes.

[D]

Notes.

[C]

I'll sit in there, appreciate the space, look around a bit, and then carry on my merry way.

[A]

You know, it's only eight miles away. The American Society of Dowsers. Oh, the national headquarters. What is a dowser? Oh, see this guy? He's holding some twigs in his hands. Actually he looks like he's holding plastic rods. But normally you would do this with, you know, like hickory twigs. And he walks around, he helps you find where to dig your. Well, his.

[C]

Oh, I'm very interested in that.

[A]

Yeah.

[D]

The miracle water work.

[A]

Yeah, well, it's, it's. It said sometimes they call it water witching and you use a device usually Kind of like fork sticks and as you walk around suddenly it starts twitching and points down at the source of the water and that's where you dig the well.

[C]

Oh, I love that. Yeah, I will go to that as well.

[A]

Yeah. 2, 000 people from all across the United States are dowsers. American Society of Dowsers has its annual convention there every year.

[C]

That's delightful.

[A]

And there's a small I love atlas next to the parking area you can use to practice your dowsing.

[C]

What a wonderful world this is.

[A]

It is. It is.

[C]

So Webster's.

[B]

Webster's is in Springfield, Mass. I think. Historic building. The American Antiquarian Society.

[A]

Are you using the Atlas Obscure to plan this trip? That's where the dog chat.

[C]

Yeah, it's one of some. The main. One of the things I look.

[A]

It's a pretty cool website.

[C]

It's a very cool website. They've got a lot of like interesting things on there. Just like strange stuff to stop by that I plug in whenever I'm. Especially a trip like this that I decided like two days in advance.

[B]

What's the destination in. In Vermont?

[C]

A bread and puppet theater which is a famous puppet museum and puppet theater and 150 year old barn that I've always wanted to go to because I've long admired their work and art and kind of their. It's like almost like a kind of neo futuristic wow approach to stuff and I don't know, seemed like the time to do it. The other pick of the week is I have got a master list of Nick Cage films online.

[B]

Have you already do any reruns?

[C]

They can be reruns if you want. I've watched. I watch Face Off. I've seen Matchstick Men so far. I need to figure out what I'm gonna watch tonight after we get done with this podcast. But I've got all of the films I think can and should count for it in this list if you want to nickvember along. It's 126 films and great. I don't know. It's a lovely endeavor.

[A]

I am. Now that I know and should I not share this but now that I know your handle on letterboxd you can show. I'll follow you because I want to know what's the best nickvember movies to see. What's. What are you watching tonight?

[C]

I don't know. I haven't decided yet. I might watch Red Rock.

[A]

I didn't even know he was in Brew Breaker.

[C]

Yeah, that he's uncredited in that. That was a bit Of a. I. I've gone through a couple different lists and I am including. I'm making some editorial judgments in my list that are even, as you should, a small, small feature. Or it's like, maybe not his participation, but it's about him in some way like that. That can count as a Nick member pick. But Brew Breaker is technically his first film appearance.

[A]

Oh.

[C]

But uncredited as. Where is he?

[A]

So you blink and you'll miss him, in other words.

[C]

Yeah. I believe he's like a guy in a car somewhere.

[A]

If I want to see the new one that he did, which. Which was kind of a parody of himself. I mean, I did see it. I really liked it.

[C]

Unbreakable or the. What's it called?

[A]

It's a funny name.

[C]

The unbearable weight of massive talent.

[A]

Yes, he. It is quite funny and good. He's making fun of himself, which is great.

[C]

I'm really excited because this is the first in November where a Nick Cage movie is being released during the month. I'm kind of scary. I'm kind of a scaredy cat. But given. Given everything, I've got to go see it in theaters. I'm gonna go see the Carpenter's Son, which releases on November 14, which is a horror movie where Nick Cage, I think, plays Jesus's uncle.

[A]

But maybe I was gonna say, if he's the carpenter's son, it's Jesus's baby boy.

[C]

Yeah.

[A]

Which very few people know about.

[B]

The.

[C]

The only thing there's no.

[A]

It's kind of the untold story of the whole thing.

[B]

Yeah.

[C]

It features Nicholas Cage as the Carpenter, so I guess he's Jesus Ska Twigs as the mother.

[B]

Carpenter would be Joseph.

[C]

Joseph.

[A]

No. Jesus was a carpenter, Wasn't he a carpenter? Yes.

[B]

Right.

[C]

Some guy named Noah Jupe as the boy. And that's it. They did. The only other details in the Wikipedia are that during filming, Cage was reportedly attacked by a swarm of bees in one of the caves intended to be a filming location.

[A]

Are Karen and Richard in the ship movie, too? I thought it was the. It was the. It was the Carpenter's biopic.

[B]

She has no idea what you're doing. Sorry.

[C]

Yeah.

[A]

Have you ever heard of the Captain and Tenille? Never mind. All right. Thank you, Paris, for those picks. Letterbox. Can we say who you are on letterboxd, or is that a secret?

[C]

Yeah, follow me on letterboxd. I'm the Void.

[A]

The Void. The Void.

[C]

I'm not. I don't. Not responsible for. I use crass language on my letterboxd. And many of the jokes will not make sense to you guys, but those are my top four favorite movies right now. All of which rule.

[A]

Jeff Jarvis, your pick of the week.

[B]

Simple 1. Since doing age verification in the UK, Pornhub's UK visitors are down 77.

[A]

So it works well.

[B]

Yeah, but this is.

[A]

Or maybe they have the same amount of visitors, but they're using VPNs and appear to be coming from somewhere other than that.

[B]

Plus they. The pornhub argues that people are going to the places that aren't trying like they are.

[A]

Yeah, that's a good point.

[B]

And that there are plenty of other.

[A]

Places you can go that don't adhere to the law.

[B]

Cyber News counted more than 10.7 million downloads of VPN apps in the UK across 2025.

[A]

Wow.

[B]

They said, where's the other number I wanted? There are something like 20. So there's some huge number of porn sites and they can't do them all. But pornhub is visible, so they're the ones who are under the thumb. It's a violation of people's freedom of.

[A]

Don't say it. Don't say it. Jeff Jarvis. He is the professor of Emeritus of Journalistic Innovation at the Craig Newmark Graduate School, the City University of New York. He's also at Montclair City and SUNY Stony Brook, author of the Gutenberg Parenthesis and magazine. Thank you, Mr. J.J. appreciate it. Appreciate you. My friend Paris Martineau is a investigative reporter at Consumer Reports, where she specializes in food safety, but covers many other topics. And she is excellent at making spooky sounds. We will see you both next week. We do Intelligent Machines every Wednesday right after Windows Weekly. That's 2pm Pacific, 5pm Eastern, 2200 UTC. Next week, the founder of Intelligent Internet, I I, Ahmad Mostak.

[C]

They finally made one.

[A]

Yeah, in a couple, actually. Is that. Is that. No, no, I think I'm wrong.

[B]

Kevin, record that.

[A]

We're recording that. Kevin Kelly's.

[B]

Kevin Kelly's next week.

[A]

Yeah. Oh, I'm excited about that. I love Kevin Kelly. Great journalist, writes for Wired, but he's been around for a long time. He was associated with Stuart, Brandon, the Whole Earth Catalog. Jimmy Wales is coming up, the founder of Wiki. Looking forward to talking to him. Dr. Anthony Vinci, the fourth intelligence revolution. Robert Seeger, Pliny the Liberator's discord moderator. He's an expert on prompt injection and jailbreaking. C.J. trowbridge from Claude. We have some great guests coming up, so we thank you so much. Yeah.

[B]

Good work, guys.

[A]

Yeah. And thanks to Jeremy Berman, for joining us. Very excited about the work he's doing at reflection AI 2pm Pacific. We'll be back next Wednesday. You can watch us live on YouTube, Twitch, Facebook, LinkedIn, X.com and Kik. Of course, if you're in the club, and I hope you are, you can watch us in the club. Twit Discord. That's a great place to hang out. Lots of fun people in there, lots of smart people. Friday is our AI user group. And of course, as Paris mentioned, don't forget, we are going to be returning to the Corn maze as we continue our D and D adventure. Paris will be there. I'll be there. Paul Thurrott, Jonathan Bennett, Jacob Ward. The horror in the cornfield is November 17th, 2pm what's the escape? Who's to say we, you know, we. We barely killed that plow and wasn't.

[C]

There like a day there was a. There was a scythe that I killed in one hit and then everybody else really struggled with the plow.

[A]

She has amazing dice skills. She's really good.

[C]

So I'm rolling those dice like nobody's business.

[B]

You're not convinced me to play games here.

[A]

Oh, it's so much fun. I. I can't wait. That's. That's coming up. We're going to do more of that stuff. That's why you join the club. Of course you get ad free versions of our shows. You're supporting the work we do, which I think is really important. But at the same time, there's a lot of fun too. TWIT tv Club Twit. If you have not yet joined Club Twit, please, we want to have you in the club after the fact on demand versions of this show. Ad support, of course, at TWiT TV Im you. You can also watch on YouTube and you could subscribe in your favorite podcast player. Audio or video or both. And if you leave us a knife's review, maybe Paris will read it next time on Intelligent Machines. Thank you everybody for being here. Have a great week. We'll see you next time. Bye bye. I'm not a human being.

[C]

Not into this animal scene. I'm an intelligent machine.