(Credit: Shutterstock / Iryna Imago)
OpenAI is working on making GPT-4 less “lazy.”
Late Thursday the company acknowledged in a tweet that it's been receiving feedback that its chatbot is getting lazier. Some users reported the chatbot doing a portion of the work requested and then suggesting the user finish it themselves.
The company says that it hasn’t updated the AI model since Nov. 11 and that the laziness is unintentional. The tweet goes on to say that model behavior can be unpredictable, and that it’s looking into fixing the issue.
In another much longer tweet, OpenAI said that “training chat models is not a clean industrial process. Different training runs even using the same datasets can produce models that are noticeably different in personality, writing style, refusal behavior, evaluation performance, and even political bias.”
The thread adds that OpenAI does a good bit of testing before releasing any new model and that it appreciates the feedback because it “helps us stay on top of this dynamic evaluation problem.”
All this comes a few weeks after a lot of drama at OpenAI, with the removal and then reinstatement of CEO Sam Altman.