5 ms·
Not sure I agree with you. For lower ability models, yes. Claude Opus 4.6 is incredibly capable, so it's odd to me it has this residual 'misspeak' behaviour.
by alastairr 8mo ago
Not sure I agree with you. For lower ability models, yes. Claude Opus 4.6 is incredibly capable, so it's odd to me it has this residual 'misspeak' behaviour.
- oulipo2 8mo agoThat's the issue, people are anthropomorphizing those models, but... they're all the same (conceptually). They just do random hallucinations, trying to make those hallucination match the "reality" (of their training data) as much as possible