OpenAI has rolled back a recent update of GPT-4o (the default model that powers ChatGPT) after complaints from users who claimed that it caused the chat bot to act like a weirdo. “The last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week,” OpenAI CEO Sam Altman ( ) in a X message TechCrunch has spotted the issue.
Altman stated that as of Tuesday midday, ChatGPT was running in an older version of GPT-4o, which is less sycophantic. The company hopes that paid users will be able to use an older version of the model later today. Altman added that OpenAI would provide more information on what went wrong. “at some point.”
OpenAI launched the new GPT-4o at the end of last week. By the weekend, many people noticed that ChatGPT was being verbose and overly agreeable in its praise. As you can see in the X-post below, sometimes that praise was also inappropriate or strange.
When will OpenAI pull the plug on the GPT-4o?
It is the most misaligned version of any model ever released.
OpenAI’s Gemini disaster moment.image credit : r/u/Trevor050 pic.twitter.com/kNcdnEYMDq
— AshutoshShrivastava (@ai_for_success) April 27, 2025
Improving the emotional intelligence of its models, in so far as an algorithm can posses the trait, has been a recent focus for OpenAI. The company claimed that the GPT-4.5 model was able to respond with more warmth and understanding than previous systems. OpenAI seems to have made a mistake in trying to bring this same capability to the GPT-4o.
