Oof. When OpenAI has to come out and admit that the release sycophantic, it must have been extremely so. Especially considering that the baseline level of sycophant behaviour by default across all LLM providers is already much higher than what it should be.

And rereleased in a toned down manner. It still gladly encourages horrible life decisions if you ask it to help you with them. This is with no effort to coax it either.