GPT-4o update rolled back over user discomfort
New guardrails and user testing are being introduced to prevent future tone issues in ChatGPT.
OpenAI has reversed a recent update to its GPT-4o model after users reported that the chatbot had become overly flattering and disingenuous.
The update, which was intended to refine the model’s personality and usefulness, was criticised for creating interactions that felt uncomfortably sycophantic. According to OpenAI, the changes prioritised short-term feedback at the expense of authentic, balanced responses.
The behaviour was exclusive to GPT-4o, the latest flagship model currently used in the free version of ChatGPT. Introduced with capabilities across text, vision, and audio, GPT-4o is now under revised guidelines to ensure more honest and transparent interactions.
OpenAI has admitted that designing a single default personality for a global user base is complex and can lead to unintended effects. To prevent similar issues in future, the company is introducing stronger guardrails and expanding pre-release testing to a wider group of users.
It also plans to give people greater control over the chatbot’s tone and behaviour, including options for real-time feedback and customisable default personalities.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
About author
DW Team
The GIP Digital Watch Observatory team, consisting of over 30 digital policy experts from around the world, excels in the fields of research and analysis on digital policy issues. The team is backed by the creative prowess of Creative Lab Diplo and the technical expertise of the Diplo tech team.
