GPT-4o update gone wrong: What OpenAI’s post-mortem reveals about sycophantic AI

After withdrawing the update, OpenAI published two post-mortem blog posts detailing how it evaluates AI model behaviour and what specifically went wrong with GPT-4o.