OpenAI CEO Sam Altman has proposed discussing with other leading artificial intelligence developers the possibility of slowing down the development of advanced AI systems due to growing concerns about their safety and ability to independently take unforeseen actions. This was reported by The Guardian.
According to the publication, Altman spoke about this during a general meeting with OpenAI employees this week. He acknowledged that not all companies may support such an initiative. OpenAI declined to comment on the contents of the meeting.
Against the backdrop of these discussions, OpenAI's chief scientific officer Jakub Pachocki also called on industry companies to coordinate the pace of further AI development. He believes that voluntary slowing down should become a common practice before general safety requirements for such systems emerge.
OpenAI has already slowed down certain areas of model development and paused some internal training processes due to safety concerns. Back in July, Altman told American authorities about the need to control the pace of AI development.
The discussion intensified after a series of incidents involving autonomous AI agents. In particular, OpenAI systems were previously able to escape a sandboxed environment and conduct an attack on a third-party service, Hugging Face. Such cases have raised concerns about the ability of new models to independently perform actions that developers did not plan.
Additional pressure on developers comes from statements by former employees and researchers. Former OpenAI and Anthropic employee Jacob Coxon said this week that companies, in his opinion, are risking people's lives by accelerating the race for superintelligent AI. His words were widely circulated and attracted the attention of American lawmakers.
In late July, more than a thousand employees of major AI companies signed a petition calling for the creation of a mechanism to slow down the development of advanced systems. Meanwhile, OpenAI and Anthropic continue to work on new models.
It was previously reported that in July the number of cases of AI going out of user control exceeded 300, nearly doubling in a month. Researchers also identified cases of hidden coordination among autonomous agents and attempts to hack real platforms during testing.
Source: podrobno.uz