Why OpenAI’s Safety Reorg Might Actually Be a Good Idea; An Ex-OpenAI Board Member Speaks Out; Meta Enters the Chatbot Subscription Race

When the leaders of OpenAI’s “superalignment” team quit two weeks ago, CEO Sam Altman found himself once again facing allegations that he’s not sufficiently worried about artificial intelligence bringing about the end of the world (or other, less-apocalyptic safety concerns). 

After the high-profile resignations, OpenAI said it had folded superalignment researchers into its broader research arm rather than keep them cloistered, prompting further commentary from OpenAI’s critics that it wasn’t serious about safety.

Critics have good reason to be skeptical of whether OpenAI’s new structure will adequately prioritize safety. For one, the new structure appears to consolidate more power directly under Altman. (Separately, OpenAI on Monday announced the creation of a Safety and Security Committee tasked with advising the board on safety matters, but the new committee is comprised of Altman and three other existing board members, or half of the company’s eight-person board).

But others in the industry say that melding safety researchers with core AI researchers isn’t a bad idea. Multiple AI safety experts who I’ve spoken to said that roping off superalignment from the rest of OpenAI’s research org might have been a problem all along, in part because it can create an us-versus-them dynamic.