Comment by simonw

That's what makes me think it's legit: the root of this whole issue was that OpenAI told GPT-4o:

  Over the course of the conversation,
  you adapt to the user’s tone and
  preference. Try to match the user’s vibe,
  tone, and generally how they
  are speaking.

https://simonwillison.net/2025/Apr/29/chatgpt-sycophancy-pro...