Dev: models were hopeless at solving puzzles a year ago, now is the time to retry
Darpinian · x · 2026-10-19
Developer Darpinian says he tried having AI models solve puzzles last year but found them 'pretty hopeless'; he believes now is the right time to revisit the idea given model progress. He also notes that not wanting to pay $10k for testing was another blocker at the time.
Related event: Developer says AI puzzle-solving was hopeless a year ago, now it's time(2 posts)→
More from Models
- Like human speech: LLMs output one word at a time while planning the whole point ahead — mlpowered · 2026-10-19
- Inside LLMs: concept representations help models infer a character's emotions to predict dialogue — mlpowered · 2026-10-19
- Interpretability research: LLMs plan many words ahead instead of just predicting the next one — mlpowered · 2026-10-19
- Only hard words left: predicting twists and conclusions forces models to build world understanding — mlpowered · 2026-10-19
- LLMs have been superhuman at next-word prediction since 2019 — only hard words remain — mlpowered · 2026-10-19
- Dev community notices Codex has 'suddenly got really good' at coding — enggirlfriend · 2026-10-19