Did Anthropic reward Opus 5.5 for saying 'let me check rather than answer from memory'?

andrey_kurenkov · x · 2026-11-05

Andrey Kurenkov observes that Opus 5.5 frequently says "Let me check rather than answer from memory," speculating that Anthropic may have rewarded this cautious behavior during training. Unverified personal observation, but an interesting window into RL-shaped model behavior.

Original post →

More from Models

Models channel →