Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Openai has Published a postmortem on the Most recent sycopian issues With the standard -ai model that is on the chattgpt, GPT-4O – Compensate that the company has forced to push back an update to the last week published.
After the GPT-4O model update, users found on social media that Chatgpt reacted excessively validating and pleasant. It was quickly become a mem. Users posted Screenshots from Chatgpt who welcome all kinds of problems, dangerous Decisions And Ideas.
In a post on X on Sunday, CEO Sam Altman recognized The problem and said that Openai would work in corrections “as soon as possible”. Two days later Altman announced The GPT-4O update was rolled back and Openaai worked on “additional corrections” to the personality of the model.
According to OpenaaiThe update, which was to make the standard personality of the model “more intuitive and effective”, was too strongly informed by “short -term feedback” and “did not completely take into account how the interactions of the user with chatgpt develop over time”.
We rolled back the GPT-4O update of the past week in Chatgpt because it was excessively flattering and pleasant. You now have access to an earlier version with more balanced behavior.
More about what happened, why it is important and how we address the sycopian: https://t.co/lohou7i7dc
– Openai (@openai) April 30, 2025
“As a result, repressed GPT -4O to respond that were excessively supportive, but insincere,” wrote Openai in a blog post. “Sycopheric interactions can be uncomfortable, causing trouble and causing distress. We are too short and work on doing it right.”
According to Openaai, several corrections are implemented, including the refinement of nuclear techniques and system requests, GPT-4O to explicitly prevent the sycopian. (System requests are the initial instructions that lead the overarching behavior and the tone of a model in interactions.) The company also builds more security lines to “increase [the model’s] Honesty and transparency “and the further evaluation of the ratings to” identify problems that go beyond the sycopian, “it says.
Openaai also says that there are opportunities that users give “real-time feedback” to “influence their interactions directly with Chatgpt” and to select from several chatt-personalities.
“[W]E’RE explores new ways to record a broader, democratic feedback in Chatgpt’s standard behavior, “wrote the company in its blog post.” We hope that the feedback will help us to reflect on a variety of cultural values around the world and understand how you want to develop chatt to develop to develop, develop […] We also believe that users should have more control over how Chatgpt behaves, and to the extent that it is safe and feasible, make adjustments if they do not agree to the standard behavior. “
