OpenAI Says It Cracked a Millennium Prize Problem. Why Is Anthropic Hitting the Brakes?

Jordi Visser (0:00)

I really do believe that Taiwan Semi not able to make enough compute, memory being in a bottleneck, the government being somewhat of a regulatory like hassle now, data centers not being built. I actually think it's good. I think the models are getting too good too fast, to be honest with you. So I think the compute delays are actually healthy. We're not ready cyber wise.

That's the thing, we're not ready.

Austin Campbell (0:22)

Hello everyone, welcome to Bits and Bips, where we explore how crypto and macro collide one basis point at a time.

We're here to discuss the latest stories in the worlds of crypto and macro. But before we dive in, remember, we have our own channels now. If you're listening on the Unchained feed, please come subscribe to those so you don't miss an episode. And before we begin, a quick break.

SPEAKER_3 (0:46)

This episode is brought to you by Oneinch Aqua, the shared liquidity layer from Oneinch. Back multiple liquidity positions with one wallet balance. And keep your tokens in your wallet until a swap fills.

See how it works at oneinch.com/aqua.

Austin Campbell (1:02)

All right. So we're doing this on September 14th, the Monday. And right now Bitcoin is in the 7,800s, ETH about 2510, gold at 4,280 an ounce. The 10 years almost at 5%, the 30 years at 5.36%, the last I looked. Brent got to around 108

Spot Bitcoin ETFs are losing a little money this time around, and everybody, by which I mean 80 to 90% of people, think the Fed will hike on Wednesday. So for our crew today, as always, I'm your host, Austin Campbell, High Scholar of Zero Knowledge Group. I'm here with my usual co-hosts, Ram Ahluwalia, Maester of Wealth, leader of Lumida, Chris Perkins, the golden hand of Franklin Crypto, joining us after a very long flight. So please stay alive, Chris. And today we have Jordi Visser, founder of Visser Labs, who is here to talk with us about our number one story, which is actually being real about five stories wedged into one, which is what's going on with AI. So let me lay this out, and then I'm gonna let you guys go to town. We've had a bunch of events within a handful of days. OpenAI said on September 8th that a group of its agents solved the Navier-Stokes Millennium Prize problem using OpenAI next generation models significantly more capable than GPT-6 Astra. They said it took 88 hours using 10,000 coordinating agents and cost 15 million.

The same thread said the model may require more deliberate choices about the pace of progress. However, later on, I will note that there are now people alleging that maybe it was not just the models. At the same time that was happening, Jacob Coxson, who spent three years in pre-training research at OpenAI and Anthropic, quit Anthropic on September 9th saying neither company is acting responsibly. They are racing to self-improving super intelligence and gambling with our lives.

He said the Navier-Stokes result is the start of a feedback loop of further acceleration and said warning shots like hugging face attacks have made pacing agreements between US labs more viable. Dario Amodei then published We Must Pace the Frontier on September 12th saying we must slow the pace at which we improve the capability of AI models.

Step one of that is Anthropic quote unilaterally committing to third-party evaluators with permanent employee level access. Steps two and three he wanted coordination among democracies with a narrow antitrust waiver and then a global regime including China. Sam Altman said I agree with Dario we need to pace the frontier and OpenAI will do the same. Notably from Anthropic, Evan Hubinger, the alignment lead said Jacob is correct here. We do earnestly believe AI could kill all humans. I personally think it is greater than 10% within the next decade.

Others did not agree.

Elon Musk on Cox's warning was seems like a set up to which Cox and replied, I'm real and these are my real beliefs. You could ask your AI researchers about me if you hadn't fired them. David Sachs told the labs, go ahead and slow yourselves. Stop pretending matter is independent when it is intertwined with Anthropics investors and staff and quote, you face massive product liability exposure if your products enable a truly damaging cyber attack. And Sachs also noted this was pleading for regulatory capture and possibly due to financial pressures. Gavin Baker said evaluators are smart as there's no section 230 style liability for model outputs on this guy, Austin Campbell from NYU, I think I know him, said Coxson was quote, a political operative who worked there for two months doing a coordinated PR tour. So the math claim was obviously contested. Tristed Buckmaster said hours before OpenAI's announcement that he and another researcher had shared results with OpenAI days earlier, Alpha J Works at Anthropic, the other researcher, and that there was an ongoing discussion between OpenAI, asking them about removing the author from Anthropic. And that this looks to the researchers like their work may have been used to inform this result. Terrence Tau said enormous technical difficulties still remain and that the Clay Institute has not yet recognized the result. Meanwhile, as always, politics. Trump in Ireland said whoever wins AI wins. I will also note Trump said that Ireland should someday be unified. I did not know Donald Trump was that type of Republican, but that's for another day. China's Global Times called the plan, quote, in essence, a Cold War playbook for the AI sector and has implied that they will not participate and disagree with the assessment about danger to humans. Even so, SoftBank closed 11% lower, SK Hynex fell, Anthropic remains and talks to raise money. And just as we were prepping for this, we found out that Donald Trump called Jensen on stage at the All In Summit and said that we will continue to accelerate. So this is a lot going on, and we are very lucky to have Jordi here right out of the start because you have been talking a lot about AI CapEx in this entire space. So I'm just going to throw it to you with this comment. What do you make of this?