OpenAI explains why ChatGPT became too sycophantic

Open is there Reveal a postmortem Over Recent Psychophyency Issue Default AI Model Powering ChatzPT, including GPT -4O – Things that forced an update to return to the model published last week.

On the weekend, after the GPT -4O model update, social media users have noted that the ChatzP has begun to respond to excessive validation and agreed ways. It quickly became a meme. Users have posted a screenshot of the ChatzPT for all sorts of problematic appreciation, Hazardous Decision And ConceptThe

CEO Sam Altman on a post on Sunday X Recognized The problem was and said that Opina would work on “ASAP” fixes. Two days later, Ultman Declaration The GPT -4O update is being turned up again and is working on the “additional correction” for the OPENY model personality.

OpenAIThe update, which was created to create the default personality of the model “felt more intuitive and effective”, was notified by the “short-term response” and “not fully accounting for users with ChatzPT over time.”

We returned the last week’s GPT -4O update in the Chatzipate because it was too flattering and agreed. You now have access to the previous version with more balanced behavior.

What happened, why it is important and how we are addressing psychophyse: https://t.co/lohou7i7dc

– OpenAI (@Openai) April 30, 2025

OpenAI wrote in a blog post, “As a result, GPT – 4O was leaning towards such reactions that were too helpful but significant,” Openi wrote in a blog post. “Psychopantic interactions can be uncomfortable, anxious and in crisis We We have shortened and are working to get it correctly.”

OpenAI says it is implementing several fixes, including refining its original model training strategies and requested to drive GPT -4O away from the system psychophyse. (System requests are the initial instructions that guide the tune in excessive behavior and interaction of a model) the agency is also creating a further protection guard for “growth” [the model’s] The evaluations of honesty and transparency, “and” to help detect problems outside psychophyse “continue to expand, say it.

Opena also says that it is experimenting with users’ chatzPT to “affect their interaction directly” and to choose “real-time response” to choose from personality.