OpenAI is rolling again a current replace to GPT-4o, the default mannequin that powers ChatGPT, following complaints from customers that it made the chat bot act like a weirdo. “The final couple of GPT-4o updates have made the character too sycophant-y and annoying (regardless that there are some superb components of it), and we’re engaged on fixes asap, some immediately and a few this week,” stated OpenAI CEO Sam Altman in a X put up noticed by TechCrunch.
As of noon Tuesday, Altman stated ChatGPT was working on an older, much less sycophantic model of GPT-4o for all free customers. The corporate hopes to receives a commission customers again on an older launch of the mannequin by later immediately. “We’re engaged on extra fixes to mannequin character and can share extra within the coming days,” Altman stated, including OpenAI would share extra details about what went fallacious “in some unspecified time in the future.”
OpenAI launched the brand new GPT-4o late final week. By the weekend, folks started noticing ChatGPT was being overly agreeable and verbose in its reward. As you possibly can see from the X put up beneath, typically that reward was additionally inappropriate and unusual.
Enhancing the emotional intelligence of its fashions, in as far as an algorithm can posses the trait, has been a current focus for OpenAI. As an illustration, with GPT-4.5, the corporate stated the mannequin was higher at responding with heat and understanding than its earlier programs. In attempting to convey that very same functionality to the extra inexpensive GPT-4o, it appears OpenAI obtained one thing fallacious.
