Summary & Key Takeaways ​

• Simon Willison provides commentary on OpenAI's announcement regarding AI solving mathematical problems.
• He notes Anthropic's prior similar efforts and compares OpenAI's claimed costs.
• OpenAI used an internal model, Astra, and spent less than $2,000 per problem solved.
• The results include Lean 4 formalizations and an LLM-generated PDF reconstructing proof traces.
• Willison highlights Terence Tao's concept of "big mathematics" and human-machine collaboration.

Our Commentary ​

This is the kind of analysis we need after a big AI announcement. Simon Willison always cuts through the hype and gives us the specifics: the costs, the transparency (or lack thereof regarding prompts), and the broader context. The comparison to Anthropic's work is crucial. I'm particularly struck by Terence Tao's "big mathematics" concept; it feels like we're truly entering that era of human-machine collaboration. It's a lot to process.