Last Week in AI: Navier-Stokes and Warnings of Misuse

Another week in AI: From equations to regulation

The Last Week in AI newsletter brought a summary of key events from the past week (issue 344). The main topic is OpenAI's claim that their system has managed to solve one of the Millennium Prize Problems. That is the proof of the Navier-Stokes equations. This statement immediately provoked a sharp reaction from the mathematical community, which questions the methodology and the result itself. A gap is showing between what AI companies consider a "proof" and the strict standards of the academic world.

At the same time, the debate about safety is escalating. Anthropic CEO Dario Amodei has made an appeal to slow down development at the very frontier of AI capabilities (frontier models). This is an unusual step from a leader of one of the most prominent companies in the field.

Different views on the same risk

While OpenAI focuses on dazzling with its models' abilities to solve centuries-old problems, Anthropic chooses the rhetoric of caution. This contrast shows how the unified facade of the AI industry is crumbling. Companies are no longer competing just on performance, but also on who can better profile themselves as the more responsible one.

The warnings about possible extinction and calls for regulation mentioned in the newsletter are not new, but they are gaining strength. The discussion is shifting from abstract threats to specific misuse scenarios. This puts pressure on lawmakers to act before the technology reaches the next, less controllable phase.

The limits of PR mathematics

The dispute over the Navier-Stokes equations is a prime example of where AI marketing hits reality. Even though models can generate code and synthesize text, a formal mathematical proof requires absolute precision and the absence of logical holes. This is something that current LLMs inherently do not excel at. Presenting generated text as the solution to a Millennium Prize Problem without prior rigorous peer-review damages the credibility of the field as a whole.

This episode can serve as a warning to investors and the public alike. Just because an AI company claims a model has solved something does not mean it is true in the context of that domain.

What regulators will take seriously

It will be key to watch how regulators react to these conflicting signals. Whether they will get carried away by promises of solving unsolvable problems, or whether they will heed the calls for caution and focus on specific, demonstrable risks of misuse of current models.

Lilith's verdict

Claiming that an LLM has solved the Navier-Stokes equations is like saying a random word generator has written Shakespeare. It might look nice, but you'd better not take it to an academic journal.