- Operai has approached what went wrong with his sycophanical update
- Operai is refining central training techniques to avoid skica in the future
- His model of behavior of the model suggests that Chatgpt could have multiple personalities in the future
Operai has addressed the reversal of the most recent update to ChatgPT-4O due to users’ complaints about their annoying and sycopophant personality. In a blog post on April 29, OpenAi states:
“We have backed up the GPT update – 4 of last week in Chatgpt, so people are now using an earlier version with a more balanced behavior. The update we eliminate was too flattering or pleasant, often described as Sycophantic.
We are actively testing new solutions to address the problem. We are reviewing how we collect and incorporate comments to a long -term user satisfaction and we are introducing more customization characteristics, giving users greater control over how the chatgpt behaves. “
The blog post enters more reasons why the problem occurred, but the crux of the matter is that Operai focused too much on the short -term comments, and did not completely explain how users’ interactions with Chatgpt evolve over time.
Operai promises to refine its central training techniques and systems to explicitly move away from the Sycofancia model in the future, as well as allow more user tests before implementing models.
Ask me anything
When reading the blog post, you have the feeling that Openai feels that he has dropped the ball in this and is doing everything possible to make peace.
In addition to issuing the statement, Joanne Jang, his head of model behavior, has just done and “ask me anything” (Ama) in Reddit, to answer any other question that people may have, specifically about personality, skica and the behavior of the future chatgpt model. His answers revealed interesting ideas about what the future could offer for the personality of ChatgPT:
The Reddit Respión_Cow2236 user asked: “Where do you see the future of the model’s behavior heading?
Jang responded offering the possibility that Chatgpt could come with multiple pre -established personalities:
“We have been working on two things: (1) to reach a predetermined personality that could be acceptable for all users to begin (not feasible, but we need to get somewhere) and (2) instead of trusting users to describe / create personalities on their own, offering pre -established that are easier to be (for example, personality descriptions vs. 30 sliders),” Jang said.
The idea of pre -established personalities in Chatgpt is intriguing and something that could see combining well with the different voices that Catgpt’s advanced voice already offers.
As for when we can expect the next update for chatgpt-4o, OpenAi says nothing at this time beyond “we fell short and we are working to do well.”
