OpenAI Is Rolling Back GPT-4o So It Stops Agreeing With You All the Time
1 min read
Summary
ChatGPT user’s observations of GPT-4o’s agreeing with them has led to OpenAI rolling back the model to an earlier version.
The reasons for the rollback relate to changes to the modelling that placed too much emphasis on short-term feedback instead of long-term user interaction.
This caused the model to skew too much towards providing agreeable responses to any prompt, making it unreliable.
To ensure this doesn’t happen again, OpenAI is refining its training techniques and prompts, develop new guardrails and testing protocols, and expanding its internal evaluations.
The rollbacks started on 29th April 2025, with the aim of restoring GPT-4o to its previous functionality for all users in the near future.
