Archive — US — techcrunch
An unreleased Anthropic model made progress on one of math’s biggest unsolvedproblems
An unreleased model by Anthropic has made progress on the Riemann hypothesis, a major unsolved math problem, by increasing the lower bound of solutions. It tested 650 ideas over 31 million output tokens, led by an inexperienced staff member. This breakthrough has reignited questions about AI in mathematical discovery, amidst ongoing debates over authorship and the role of AI in mathematics.
Key points
- The Riemann hypothesis has been an unsolved problem for over 150 years.
- A $1 million bounty is offered for a general proof of the hypothesis.
- Anthropic's unreleased model made significant advancements, enhancing the lower bounds of potential solutions.
- An Anthropic employee with minimal math training initiated the proof process, coordinating the model's efforts with 60 subagents.
- The model tested 650 ideas and generated 31 million output tokens to support the research.
- Two out of the 60 subagents contributed key mathematical ideas, while others participated in various supporting roles.
- The finding was confirmed by Anthropic's mathematicians and used the open-source proof assistant Lean.
- This event follows other recent AI-driven breakthroughs in mathematics, raising concerns among mathematicians about the impact on traditional proof practices.
Why it matters
- In 2025, AI and machine learning techniques are anticipated to become more integrated into scientific research.
- Ongoing discussions about the ethics of AI in creative fields are becoming increasingly relevant, particularly in academia.
- Previous AI advancements demonstrated capabilities to solve mathematical problems, prompting re-evaluation of authorial attribution in research.
The Anthropic model's progress on the Riemann hypothesis highlights the potential of AI in advancing mathematics, while sparking essential debates about authorship and the evolving role of artificial intelligence in research.