logoalt Hacker News

SomaticPirateyesterday at 10:39 PM0 repliesview on HN

Are these benchmarks correct that adding Anthropic's Constitutional AI system prompt lowered results across all the models?