Breakthrough Achieved by Unveiled Anthropic Model on a Major Mathematical Conundrum
The Riemann hypothesis has been a central enigma in mathematics for over 150 years, regarding the distribution of prime numbers. Currently, a $1 million prize is available for a verified general proof, which remains unsolved.
While modern AI models haven’t cracked the hypothesis yet, they are making surprising strides, prompting discussions about whether these technologies can unearth new scientific and mathematical insights.
Recently, Anthropic revealed that a forthcoming model achieved notable advancements related to the Riemann hypothesis, which enhanced the lower limits of solutions to the conjecture.
What’s particularly remarkable is how this progress was accomplished: an Anthropic team member with limited mathematical expertise encouraged the model to attempt a proof, allowing it to orchestrate its work over the next 36 hours.
In total, the model examined 650 ideas for potential solutions, utilizing 60 sub-agents and expending 31 million resources.
According to a note in the research paper, “Out of the 60 subagents, two focused on developing key mathematical concepts; 13 offered ideas to these agents, 30 tried (and failed) to create new ideas, 13 acted as validators to ensure the correctness of the arguments, and the last two assisted in drafting the initial paper.”
This finding was substantiated by two mathematicians from Anthropic and formalized with the help of the open-source proof assistant known as Lean.
This achievement is part of a series of mathematical advancements driven by Large Language Models. Throughout the year, AI models have successfully addressed several problems related to Erdős, with recently released more powerful models yielding even greater results. OpenAI unveiled ten significant results confirmed by its internal “Astra” model, while another team at Anthropic disproved the long-held Jacobian conjecture.
The increasing number of results has generated both enthusiasm and trepidation within the mathematical community. In June, a group of leading mathematicians publicly expressed concerns that AI could threaten key principles in the field, particularly regarding the attribution of true mathematical proofs to specific authors who can take responsibility for their validity.
However, the mathematics community remains divided on how to adapt to these new research methods. In response to the declaration, Fields Medal laureate Timothy Gowers suggested in a blog post that AI’s impact might lead to a more intricate and beneficial transformation in mathematics.
“If we find ourselves in a situation where mathematical theorems are no longer tied to specific mathematicians, perhaps that won’t be any more troubling than the fact that stars aren’t named after astronomers, and many have no names at all,” Gowers remarked.
