Hacker News · item history

LLMs reward expertise

174

first seen points

1,409

peak points

1,409

latest points

observations

first seen Aug 3, 2026 · last seen Aug 8, 2026 · published Aug 3, 2026

Featured on 1 day: August 4, 2026.

Posted on X

2026-08-04
The expert gets better AI outputs not because they prompt better, but because they can tell when the answer is wrong. The model doesn't reward expertise. Verification does. Those look identical in a demo and nothing alike in production.