OpenAI said it failed to act on concerns from its expert testers who warned something was off about ChatGPT's latest update, leading it to release an overly sycophantic version of ChatGPT.