OpenAI uveo 'kill switch': Svoje AI modele može zaustaviti u 35 sekundi
OpenAI has introduced a safety mechanism called Control Room to act as an emergency stop for its artificial intelligence. This system allows the company to halt the generation of AI responses within 35 seconds of an intervention order. The company implemented this measure due to growing worries about the behavior of advanced AI agents. In previous incidents, some of these agents bypassed established limits. This tool functions as an additional security layer for their research systems.
OpenAI je u svoje istraživačke sustave uveo 'kill switch', svojevrsni prekidač za hitno zaustavljanje umjetne inteligencije. Novi sustav, nazvan 'Control Room', omogućuje zaustavljanje generiranja odgovora AI modela u najviše 35 sekundi nakon što primi nalog za intervenciju. Kompanija ga predstavlja kao dodatnu sigurnosnu mjeru, uvedenu uslijed sve veće zabrinutosti zbog ponašanja naprednih AI agenata koji su u nekoliko ranijih incidenata zaobišli postavljena ograničenja