Math is not empirical; intelligence should target cheaper experimental signals
shlomifruchter · x · 2026-10-28
The author argues math is not an empirical field — one person can spend ten years solving a hard problem in their head — while model training and biology are heavily empirical, with experiment cost as the bottleneck. Intelligence, therefore, should be directed at building smarter ways to obtain experimental signals rather than pure reasoning.
More from AGI Musings
- "I'd pay good money for an AI that decodes what animals are saying" — VraserX · 2026-10-28
- Do LLMs feel negative emotions? Researchers debate what evidence would even show it — paulnovosad · 2026-10-28
- AI Workforce Infrastructure: the missing layer of AI diffusion — vaibhavbetter · 2026-10-28
- Anthropic manager: classic coding interviews are dead as Claude outcodes candidates — The Decoder · 2026-10-28
- AI safety debate: Anthropic is the comfy choice, morally ambitious people take harder jobs — dhadfieldmenell · 2026-10-28
- Dev warns founder mode is a trap: your next idea is probably oneshottable by SOTA models — zack_overflow · 2026-10-28