Still, the models are improving much faster than the efforts to understand them. And the Anthropic team admits that as AI agents proliferate, the theoretical criminality of the lab grows ever closer to reality. If we don\u2019t crack the black box, it might crack us.<\/p>\n
\u201cMost of my life has been focused on trying to do things I believe are important. When I was 18, I dropped out of university to support a friend accused of terrorism, because I believe it\u2019s most important to support people when others don\u2019t. When he was found innocent, I noticed that deep learning was going to affect society, and dedicated myself to figuring out how humans could understand neural networks. I\u2019ve spent the last decade working on that because I think it could be one of the keys to making AI safe.\u201d<\/p>\n
So begins Chris Olah\u2019s \u201cdate me doc,\u201d which he posted on Twitter in 2022. He\u2019s no longer single, but the doc<\/a> remains on his Github site \u201csince it was an important document for me,\u201d he writes.<\/p>\n Olah\u2019s description leaves out a few things, including that despite not earning a university degree he\u2019s an Anthropic cofounder. A less significant omission is that he received a Thiel Fellowship, which bestows $100,000 on talented dropouts. \u201cIt gave me a lot of flexibility to focus on whatever I thought was important,\u201d he told me in a 2024 interview. Spurred by reading articles in WIRED, among other things, he tried building 3D printers. \u201cAt 19, one doesn\u2019t necessarily have the best taste,\u201d he admitted. Then, in 2013, he attended a seminar series on deep learning and was galvanized. He left the sessions with a question that no one else seemed to be asking: What\u2019s going on in those systems?<\/p>\n Olah had difficulty interesting others in the question. When he joined Google Brain as an intern in 2014, he worked on a strange product called Deep Dream<\/a>, an early experiment in AI image generation. The neural net produced bizarre, psychedelic patterns, almost as if the software was on drugs. \u201cWe didn\u2019t understand the results,\u201d says Olah. \u201cBut one thing they did show is that there\u2019s a lot of structure inside neural networks.\u201d At least some elements, he concluded, could be understood.<\/p>\n