Recently, xAI engineer Hieu Pham made a "bombshell" announcement on social media, claiming that xAI's large language model Grok3 successfully proved the Riemann Hypothesis, leading to a suspension of the model's training. This news quickly sparked heated discussions in the AI community, with netizens expressing shock and struggling to discern the truth. After all, the Riemann Hypothesis, as one of the seven Millennium Prize Problems, is hailed as the "crown jewel of conjectures," and its difficulty is unimaginable.

image.png

Hours later, Pham revealed the truth in another post: it was just a joke. This "mix-up" originated from a claim by netizen Andrew Curran, who stated that Grok3 encountered a "catastrophic event" during its training.

image.png

In response to the increasingly absurd rumors, xAI co-founder Greg Yang couldn't help but post sarcastically, "Yes, yes, Grok3 just started attacking the security guards in the office while training." Another researcher, Heinrich Kuttler, humorously stated, "The situation was terrible! We later replaced all the bad weights with nan (Not a Number) to recover." Netizens also joined in on the joke.

image.png

Although this "farce" ended in laughter, it also sparked reflections on AI's mathematical capabilities.

So, how far is AI from solving a Millennium Prize Problem like the Riemann Hypothesis?

We can gain some insight from the performance of AlphaProof, an AI math proof tool developed by Google's DeepMind team. AlphaProof successfully solved three problems in the 2024 International Mathematical Olympiad (IMO), with the sixth problem being dubbed the "ultimate boss" due to its extreme difficulty. AlphaProof demonstrated strong logical reasoning and creative thinking during the problem-solving process; for instance, in the second problem, it cleverly chose to consider the number ab+1 to construct its proof, a strategy aligned with human problem-solving approaches.

image.png

While AlphaProof achieved remarkable results, AI still has a long way to go to tackle top mathematical challenges like the Riemann Hypothesis. Since its proposal in 1859, the Riemann Hypothesis has a history of 165 years, with countless mathematicians dedicating their efforts to it, yet a complete proof remains elusive.

For AI to prove the Riemann Hypothesis, it needs powerful computational capabilities and deep reasoning skills. Currently, AI can find provable theorems by exhaustively searching all possible proofs, but this requires astronomical computational resources. Moreover, AI must also possess the ability to understand and apply existing mathematical tools to play a more significant role in mathematical research.

Some AI experts predict that by the end of 2026, AI will become "super mathematicians," capable of solving challenges like the Riemann Hypothesis. Elon Musk has also promised that Grok3, trained with 200,000 H100 units, will be released by the end of the year, delivering astonishing performance.

In the future, whether AI can achieve breakthrough advancements in the field of mathematics remains to be seen.