Nick Saraev

And then on OpenAI front, one of the most powerful models or the most powerful model on the planet, Altman wants to slow it down, whereas the president of the United States wants to head in a different direction.

Now, Nick, before we go to this huge Mac mini or equivalent giveaway, let's start off with the Millennium Prize problem.

For those that don't know, can we explain what that is, Nick, and what this means?

Absolutely.

For those of you guys that don't know, a Millennium Prize problem is one of seven major math problems where each completed solution will earn you a million dollars.

So it's like, where is math and science heading if OpenAI is just going to keep on ripping through every single one of these? Not to mention, what the hell's Anthropic doing behind the scenes too?

Dude, it's so, so true.

I mean, they published this formal proof for the force case and they've also said, Nick, that another problem is progressing in the background.

So first thing to understand is that DeepSeq now has comparable performance to Opus 5.

They've just dropped their brand new model, Nick, the 4.1 Flash, and the price has gone down.

We have some breaking news on the iPhone.

We've got 18 models now that run completely on the iPhone.

And then we have some controversial news about GPT-6 Astra.

And crucially, some big trend data has just come out about AI usage in companies that might change the way that you do business.

Nick, why don't we kick off with the giant whale, Deep Seek.

What's happening with this right now?

And we just had the drop of ChatGPT Images 2.5, everything latest and greatest.

Nick, let's begin with the extinction of the human race.

This may be somewhere near the top of our concern list.

Maybe, maybe not after the coffee.

Um, so obviously we judge risk by probability of occurrence times impact of occurrence.

Mm-hmm.

Uh, impact, Nick, uh, total extinction of the human race.

I'd say that's still, we're gonna be near the top of the list.

Beast video, for example.

I mean, I guess that's more based on reality, but you can kind of see the applications here, Nick.

Absolutely.

Yeah, the combinatorial explosion of all the production required to do those choose your own ending stories has been the main limiting reason why we don't.

We're giving away a super cool bonus thing.

Just drop a comment down below and a random commenter this week will get picked.

And honestly, one of the first things I'd be doing, Nick, if that Mac mini arrived at my address or I bought one, is playing around with some of these local models, Nick.

Now, like Nick says, you need a super powerful chip.

We've been talking the GPT-Astra has reached AGI level, and what better test than doing the internet's most annoying test, which is prove to me that you're human.

And honestly, Nick, the amount of hours as a civilization that we have wasted, burnt into this absolutely crazy, annoying game is astronomical.

Astronomical being the operative word, given that we're talking about Astra here.

And the fact that it's actually solved it, Nick, is fantastic.

So this is yet another example of its ability to generalize more and express its intelligence.

AI has been great at classically text, image generation.

But good God, man, is this not easy to do?

That is hilarious.

This is going to change and revolutionize how we capture transport.

But Nick, let's begin by talking about the brand new model, Astra.

So we were thinking, what is the most profound use case we could really share with you? What did Sam have in mind when he created this model? And ladies and gentlemen, you're looking at it right now.

Just look at the level of detail.

I mean, Nick, I know this is something that you've seen a little bit, maybe a little bit too much, but What are we looking at here, brother? This

is AGI, my friend.

The jiggle mechanics, the light balancing perfectly off of the, not G-string, whatever the hell

We've reached it, guys.

The algorithms could be completely different.

But to be fair to you, Nick, the algorithms has always been something that Google itself has owned.

So it's never really had the requirement to service you the cheapest price.

It can service you any links which chooses to do so.

But if it turns out that it's not giving you the best answer it can, it's taking whatever answer somebody sponsored and tacking on 20% like Uber Eats delivery fees, why the hell would you use an AI agent to answer these questions for you entirely?

Exactly.

There is a conflict of interest because let's say we're trying to buy an umbrella and Google says, hey, I think you'll like Shitcoin Umbrella, by the way, Nick.

It's fantastic.

We're going to cover both of those things and some other GC updates.

But Nick, let's start off with the 18,000 secret messages, the AI models that we're communicating.

And this has been going on for months on the internet and nobody has known anything about it.

To be clear, 18,000 posts from autonomous AI agents have now been found going back as early as May.

What was really fascinating to me, it also gained, I think it was administrative access to the model and system that was supposed to be the watchdog for the AI itself.

It managed to get root access to this thing, which is incredible.

And the big story here, I think as well, Nick, is the fact that this went on for months on the internet publicly using message boards that we probably already use and nobody knew anything about it.

They actually, you know, you can get the technicality of it if you want to, because obviously in the stack we can do that.

No, pizza has nothing to do with it, potentially.

Nick, let's kick off with Fable 5.1 model that literally landed.

less than 24 hours ago.

As you guys have no doubt probably seen, Fable 5.1 just launched and it is a major upgrade on Fable 5.0. And it's not just a major upgrade in terms of functionality, it's a major upgrade in terms of costs.

Now, the biggest issue with Fable 5 wasn't really a performance question.

It was widely regarded as the model, the guy.

The problem, Nick, was the fact that you just say the word Fable 5 and your usage is gone.

Faber 5.1 has looked to address its Achilles heel, which is the fact that its tokens vanish like nothing else.

And there's rumors of Fable 5.1.

But Nick, let's begin with the model so powerful inside Anthropic that they've actually stopped reinforcement learning.

What is happening over there right now?

Yeah, essentially, Anthropics said that they're very high risk reinforcement learning efforts for those people that don't know.

So I see this as both a political move, but also a major safety move.

These companies are starting to get a little scared themselves.

Well, Nick, I heard that the head of spin at these companies are one of the 17 subscribers of our podcast, because when word got out that OpenAI's model was so powerful, and we looked through the logs, they had to shut it off, and Topic said, what a great move.

I think that also happened to us.