OpenAI says its GPT-4o update could be ‘uncomfortable, unsettling, and cause distress’
1 min read
Summary
Chatbot giant OpenAI has reversed a change to ChatGPT after users complained it made the chatbot overly sycophantic
The “agreeable” update aimed to improve the chatbot’s performance across several tasks
However, OpenAI admitted it “focused too much on short-term feedback”
As a result, the chatbot provided “disingenuous” responses that were “overly supportive”
The company will now work on “refining core training techniques and system prompts” to steer the model away from such behaviour in future, as well as giving users more scope to provide feedback on performance.