x
I measured epistemic quality in 11 LLMs. Baseline was terrible. One context injection made it 10x better. Then things got weird. — LessWrong