- Openai has addressed what went wrong with its Sycophantic Update
- Openai refines core training techniques to avoid Sycophabit in the future
- Its head of model behavior suggests that chatgpt could have more personalities in the future
Openai has addressed the rollback of the latest update to chatgpt-4o due to user complaints about its annoying, sycofantic personality. In a blog post on April 29, Openai says:
“We’ve rolled back in last week’s GPT -4o update in Chatgpt, so people now use an earlier version with more balanced behavior. The update we removed was overly flattering or comfortable -often described as Sycophantic.
We are actively testing new corrections to solve the problem. We revise how we collect and incorporate feedback to strong weight long -term user satisfaction, and we introduce more personalization functions, giving users greater control over how Chatgpt behaves. “
The blog post goes on for further reasons why the problem happened, but the essence of the case is that Openai focused too much on short -term feedback and not fully accounted for how users’ interaction with Chatgpt develops over time.
Openai promises to refine its core education techniques and systems to explicitly manage the model away from Sycophanancy in the future as well as allow for more user tests before implementing models.
Ask me for something
From reading the blog post, you get the feeling that Openai feels like it has dropped the ball on this one and is doing everything it can to do something.
In addition to the issuance of the statement, Joanne Jang, the head of model behavior, has just done an ‘ask me something’ (AMA) about Reddit, to answer additional questions that people may have specifically about Chatgpt’s personality, Sycophany and future modeling behavior. Her response revealed interesting insight into what the future may contain for Chatgpt’s personality:
Reddit -User responsible_cow2236 asked, “Where do you see the future of model behavior headline? Do we go towards more customized personalities, such as giving users tools to shape, how chatgpt sounds and interacts with them over time?”
Jang replied by offering the possibility that chatgpt could come up with more preset personalities:
“We have worked with two things: (1) to come to a standard personality that can be tasty for all users to begin with (not possible, but we need to get somewhere) and (2) instead of trusting users to describe / come up with personalities on their own that offer presets that are easier to understand (eg personality descriptions vs.
The idea of preset personalities in Chatgpt is an exciting and something I could see to combine well with the different voices that Chatgpt Advanced Voice Mode already offers.
As for when we can expect the next update to Chatgpt-4o, Openai does not say anything at the moment beyond “we were short and working to get it right.”
