Use an LLM to rank genuinely unsolved problems, not the ones already solved in training data
p_e_cooper · x · 2026-07-27
The poster argues that if the goal is to use an LLM on an unsolved problem with limited qualifications, the better strategy is to have it survey a broad set of genuinely unsolved problems and choose the one it is most likely to crack.
The reply adds a sharper point: the “easy problems” are already in the training data, so the useful frontier is not obvious textbook tasks but problems that remain open in the real world.
Related event: Attempt to Solve P vs NP with LLMs Sparks Mockery and Debate(4 posts)→
More from AGI Musings
- Steven Pinker says intelligence is bounded, not a limitless scalar like height — SydSteyerhart · 2026-07-27
- A poster imagines AI evolving into synthetic, collective, and beyond-intelligence systems — CurieuxExplorer · 2026-07-27
- AI surpassing human intelligence will be noticeable, as Claude's cryptic output proves — zetalyrae · 2026-07-27
- AI lab staff have gone strangely quiet about next-year capability predictions — ChrisGPT · 2026-07-27
- AI could erode science by flooding research with credible slop — rbhar90 · 2026-07-27
- Organizations may already be the planet’s superintelligences — eldonredwards · 2026-07-27