Kids as raw chains of thought: random exploration, backtracking, few induction steps
giffmana · x · 2026-10-14
giffmana (Andrej Karpathy) quips that 5-6 year olds are great because you can monitor their raw, unfiltered chain of thought all day: articulate enough to speak it, but not yet filtering it. Lots of random exploration, backtracking and forgetting, with no more than 2-4 induction steps — an analogy for LLM reasoning behavior.
More from Fun
- Building Blender scenes entirely with AI: daytime lighting and fog handled far better — jongranskog · 2026-10-14
- Snark: Anthropic's 'danger marketing' forced the whole industry to follow suit — StewartalsopIII · 2026-10-14
- Hook a $3,000 dental scanner to AI for daily brushing coaching — what's it worth? — NathanpmYoung · 2026-10-14
- Tristan Harris quotes Nick Land's 'Meltdown' on Tucker Carlson: capitalism is an AI from the future — SydSteyerhart · 2026-10-14
- Zuckerberg podcast clip sparks the 'that's my quant' meme in AI circles — eter_inquirer · 2026-10-14
- DeLorean joins Airspeeder International Series, first American team debuting in November — evilrabbit_ · 2026-10-14