Hacker News Reader

I recognize the sarcasm. The data I can find says it's performing at baseline however?

Yeah, that's my point. Humans are not reliable LLM evaluators. "Secret model nerfs" happen in "vibes" far more often than they do in any reality.