OpenAI's AI solves unit distance problem, Karpathy joins Anthropic for RSI
AI creates new math knowledge; Karpathy aims for recursive self-improvement at Anthropic.
OpenAI has delivered what's being called the first genuinely impressive mathematical result from an AI: a solution to the unit distance problem. This marks a significant leap in AI-generated knowledge, going beyond pattern matching to create new proofs. The breakthrough suggests AI is beginning to contribute original insights in pure mathematics, a domain long considered a bastion of human intelligence.
In parallel, Andrej Karpathy has joined Anthropic to work on recursive self-improvement (RSI), aiming to create AI systems that enhance their own capabilities. He plans to return to education work afterward, but acknowledges that success could make that return unnecessary. Separately, METR's risk report on frontier models concluded that while current systems don't yet pose existential threats due to insufficient capability and opportunity, that window is closing. Elon Musk's lawsuit against OpenAI was dismissed for being filed too late, and the White House is expected to issue a new executive order on AI today.
- OpenAI solved the unit distance problem, the first AI-generated math result widely considered impressive by mathematicians.
- Andrej Karpathy joins Anthropic to lead recursive self-improvement efforts, potentially creating self-improving AI.
- METR risk report says frontier models lack means, motive, and opportunity for catastrophe—but the situation may change soon.
Why It Matters
AI is transitioning from tool to creator of new knowledge, raising stakes for safety and alignment work.