Computer Science
There is no correct answer to \"write a good summary\". So models are not trained to be good — they are trained to be rated highly
18 Sept 2026 · 8 min read
Predicting the next word has a right answer in the data. Being helpful does not. The industry's solution was to stop defining quality and learn it from human comparisons — which works, and produces every strange habit these systems have.
Read more →
