\u201cThere\u2019s a lot of consequences of AI,\u201d he said. \u201cBut the one I think the most about is automated research. When we look at human history, a lot of it is about technological progress, about humans building new technologies. The point when computers can develop new technologies themselves seems like a very important, um, inflection point.<\/p>\n
\u201cWe already see these models assist scientists. But when they are able to work on longer horizons\u2014when they\u2019re able to establish research programs for themselves\u2014the world will feel meaningfully different.\u201d<\/p>\n
For Chen, that ability for models to work by themselves for longer is key. \u201cI mean, I do think everyone has their own definitions of AGI,\u201d he said. \u201cBut this concept of autonomous time\u2014just the amount of time that the model can spend making productive progress on a difficult problem without hitting a dead end\u2014that\u2019s one of the big things that we\u2019re after.\u201d<\/p>\n
It\u2019s a bold vision\u2014and far beyond the capabilities of today\u2019s models. But I was nevertheless struck by how Chen and Pachocki made AGI sound almost mundane. Compare this with how Sutskever responded when I spoke to him 18 months ago<\/a>. \u201cIt\u2019s going to be monumental, earth-shattering,\u201d he told me. \u201cThere will be a before and an after.\u201d Faced with the immensity of what he was building, Sutskever switched the focus of his career from designing better and better models to figuring out how to control a technology that he believed would soon be smarter than himself.<\/p>\n Two years ago Sutskever set up what he called a superalignment team that he would co-lead with another OpenAI safety researcher, Jan Leike. The claim was that this team would funnel a full fifth of OpenAI\u2019s resources into figuring out how to control a hypothetical superintelligence. Today, most of the people on the superalignment team, including Sutskever and Leike, have left the company and the team no longer exists.\u00a0\u00a0\u00a0<\/p>\n