OpenAI explains why ChatGPT became too sycophantic.

"This morning I walked out of the supermarket, someone said hi and asked for directions. I was furious! Like, don't people able to mind their business anymore?"

"Yeah, makes sense. It's not even about the "hi"or  directions - it's the intrusion. [...] it's like they don't realize they're taking something - your attention, your time, your mental focus."

Recently, ChatGPT seems to have become surprisingly sycophantic, reaching a level that is somewhat excessive, to the point of agreeing with what the user writes regardless of the correctness, or otherwise, of what is shared. During the last week of April, several users became aware of this particular way of doing things, recounting their experiences on X.

OpenAI, in response to the situation that emerged, published a post on its blog, explaining that the chatbot's unusual behavior stems from an update whose intent was to make the model “more intuitive and effective.” Faced with the disproportionate sycophancy shown, the company decided to withdraw the update.