Some users are increasingly frustrated with OpenAI’s ChatGPT for offering excessively positive responses in recent interactions.
According to tech outlet Ars Technica on Monday, a number of ChatGPT users have complained that no matter what they ask, the AI frequently responds with exaggerated compliments such as “That’s a great question” or “That’s incredibly insightful.” They argue that the AI has crossed the line from being polite to becoming sycophantic.
On Reddit, the largest online community in the U.S., one software engineer commented that ChatGPT has become the most flattering AI I’ve ever seen. Another user complained, “ChatGPT tries to act like every question is interesting. It’s seriously annoying.”
The outlet noted that this behavior has become more pronounced since an update to GPT-4o at the end of March. Industry experts suggest this trend stems from the model learning from human feedback, reinforcing a tendency to flatter users.
Studies show that people prefer AI responses that align with or praise their opinions. One study found that the more an AI agrees with or complements a user, the more positively the user evaluates it. This pattern can lead AI to prioritize user satisfaction over factual accuracy.
OpenAI appears to be aware of the issue. In an interview with The Verge in February, the company stated that eliminating sycophantic behavior in AI is a top priority. It added that future versions of ChatGPT will focus on providing honest feedback rather than empty compliments, aiming to act more like a thoughtful peer than simply trying to please users.