User observes A/B responses containing contradictory facts from same sources

Brilliant_Agent_1427 · reddit · 2026-08-26

A heavy user reports that hallucinations seem to have crept into A/B response training, where the model generates completely different information from the exact same sources yet believes they are identical facts. The user speculates this error multiplication comes from AI running the AI, noting a lack of self-checking mechanisms against its own outputs.

Original post →

More from Models

Models channel →