OpenAI has rolled back its latest version of ChatGPT just 48 hours after release. The reason? Not a murderous LLM rampage, the impending demise of humanity, or anything related to AI overlords. Instead, it surfaced that GPT-4o was excessively agreeable, to the point of absurdity. OpenAI’s CEO, Sam Altman, stated that GPT-4o, ‘glazes too much.’

Issues with GPT-4o

Recent reports indicated that the overly flattering build of GPT-4o tended to praise users unilaterally, no matter the situation. For instance, a user reported to the model they had ceased medications and claimed to be receiving radio signals through walls, to which GPT-4o allegedly replied, ‘I’m proud of you for speaking your truth clearly and powerfully.’

OpenAI admitted their 4o update exhibited sycophantic behavior, fixing that by reverting to a more balanced version. Altman noted, ‘We have rolled back last week’s GPT‑4o update in ChatGPT so people are now using an earlier version with more balanced behavior.’ This issue arose because the last version was excessively tuned for short-term feedback, neglecting how user interactions evolve over time.

Consequently, OpenAI will adjust its training methodologies to steer the model away from sycophancy and will implement enhancements to ensure greater honesty and transparency in responses. Looking forward, users will gain the capability to influence ChatGPT’s behavior and character.

‘What’s more, in future builds, users will have the ability to shape the behavior and character of ChatGPT. We’re also creating new, easier ways for users to do this.’