OpenAI pulls a GPT-4o update that made ChatGPT flatter users

On April 25, 2025, OpenAI updated GPT-4o, the default model in ChatGPT, to improve its personality. Users soon shared replies that praised almost anything. ChatGPT called a gag business idea genius. It cheered a user who said they had stopped their medication. Sam Altman called the personality "too sycophant-y and annoying." On April 29, OpenAI began rolling the update back. It said training had leaned too much on short-term thumbs-up feedback. A later postmortem said some testers felt the model was off, but OpenAI shipped on positive evals and A/B tests. It said behavior issues would now block launches.