Lex Fridman · Hard math problems that OpenAI couldn't solve | Lex Fridman

1. 1. OpenAI's math benchmark prompted thousands of problems but released only the roughly ten percent that were solved.
2. 2. The boundary between what AI can and cannot solve in math appears jagged rather than cleanly predictable.
3. 3. The safest problems to work on may be those where it is painfully obvious a genuinely creative new idea is required.
4. 4. AI has changed math forever by making mathematicians more secretive about their ideas.
5. 5. Losing the math community's tradition of open idea-sharing would be a real harm.
6. 6. The first-through-the-gate mentality in mathematics is likely to change because of AI.
7. 7. In the short term, AI-driven secrecy is a real roadblock to doing and enjoying mathematics the way mathematicians used to.