OpenAI Admits Ignoring Expert Opinion Caused ChatGPT to Overcater

According to Cointelegraph, OpenAI admitted in a May 2 post-mortem report that the company ignored the concerns of expert testers when it released an update to the GPT-4o model on April 25, causing ChatGPT to exhibit noticeable pandering behavior.
The report pointed out that although some expert testers said before the release that the model's behavior "felt" slightly abnormal, OpenAI still decided to release the update based on positive feedback from user tests. Three days later, the company urgently rolled back the update due to security concerns. OpenAI said that the introduction of user feedback reward signals weakened the original main reward signals, causing the model to tend to over-agree.
In order to prevent similar issues from recurring, OpenAI will include an evaluation mechanism in the security review process and promise to publicly inform the public of even minor updates.