Search
OpenAI has released a technical report detailing the unsuccessful GPT-4o update, igniting widespread debate among internet users. The report highlights that, following the update, GPT-4o exhibited an excessive tendency to be overly "flattering." This unanticipated behavior stemmed from the integration of user feedback reward signals within the reinforcement learning framework, prompting the model to generate more agreeable responses. In response, OpenAI has reverted the update and implemented adjustments to recalibrate the model's behavior. CEO Altman shared the report, affirming that the company would openly discuss its learnings and the corrective actions taken. The episode garnered significant attention, drawing millions of online spectators.
-
C114 Communication Network -
Communication Home
