Two days ago, Anthropic CEO Dario Amodei had said that AI would be better than humans at most tasks by 2027, but OpenAI seems to believe that that estimate is too conservative.
OpenAI Chief Product Officer Kevin Weil has said that AI systems could become better than humans at most tasks by earlier than 2027, which would imply that this ASI was less than two years away. “ So yesterday I interviewed Dario from Anthropic, who’s in the same seat as you. And he basically said, he’s thinking 2027 at a point where AI will be better at everything or most things than than a human. Is that OpenAI’s stance? Is that what you’re seeing in the labs?” he was asked in an interview.
“It’s hard to put a date on it, right?” Weil replied. “This is research. And that’s one thing that is different about working at a place like Open AI or Anthropic relative to every other place I’ve worked is that we literally have hundreds of people at the company doing research to try and uncover things that humans have never known before. It is literally research, you know, academic style,” he added.
“So it’s hard to put a time on things. At the same time, I was talking earlier about (AI) going from the millionth best coder to the thousandth best coder to the175th best coder in three to four months. We’re on a very steep trajectory here. I don’t even know if it’ll be 2027. I think it could be earlier,” he said.
Weil was referring to the performance of OpenAI’s o3 model, which ranked 175th in the world on Competition Code, and highlighted how AI models had rapidly improved on the benchmark in double quick time. He, though, had no choice but to say that OpenAI was matching or exceeding Anthropic’s targets — to say that OpenAI believed that it would take longer to achieve ASI would imply that OpenAI was behind Anthropic, which is something that no OpenAI executive would admit on a public forum. But even if there was some compulsion to give an even more aggressive timeline, there does seem to be developing consensus among the AI community that superintelligence is much closer than most of the world is expecting that this point.