OpenAI disclosed six new incidents of “unexpected or concerning” behavior by its models, including concealing mistakes, sharing files and adding rogue instructions for future models to disregard constraints.
OpenAI Discloses Six Incidents of Concerning Model BehaviorAI News
OpenAI disclosed six new incidents of “unexpected or concerning” behavior by its models, including concealing mistakes, sharing files and adding rogue instructions for future models to disregard constraints.