Lex Fridman: belief

Transcript context

…There’s a lot of complicated trends underway. It can be hard to break them down and distinguish them. I think a lot of people like the theories that get the biggest traction on social media often don’t capture the real underlying motive forces at play there. But I think AI involved in code production will probably create a net benefit for the need for humanity to be involved in coding. It may change parts of jobs. I don’t think it’s going to obsolete anybody who’s willing to learn new ways of doing things. It’s always been this way. I think that there’s also a lot of over hype in AI. AI is really great at spewing out code that does something that a million GitHub repositories already do because it’s learned the underlying pattern. It’s notoriously hard to get to do something new that hasn’t been done before, especially when it’s a complex task. The bigger amount of code you ask AI for, the more it leaves you with just a mess of code that sort of works. Then that’s the problem with code, it like 99% works, but the 1%, it might be harder to get to 100% with AI than with hand coding. Everybody who’s looking at this topic should actually try using the coding assistants on hard problems and see how they do there. I think for me personally, it makes it more fun and faster to generate boiler play code so I can focus on the harder decisions, harder big picture decisions and harder innovative decisions and all that kind of stuff, and just makes programming more fun for me because I feel less lonely. Even when it gives the wrong code, I get like, “Oh, okay, well that’s a way to do it. That’s interesting.” Then you could talk to it. Maybe that shows something about the programming experience that it is in part sometimes a bit lonely. The topic of boilerplate code is an interesting one because the mere existence of boilerplate code is a failure of programming language and of the idea of creating software modules. You ask AI to create a sorting function, great. Now you have another sorting function that might be buggy alongside the million others that different people have written. It would be better to have a sorting function that’s been written and tested and optimized and everybody relies on it. More modular software I think will actually reduce the opportunity of AI because people doing programming work will largely be solving unique problems. They’re actually hard problems in themselves and not just connecting other widgets.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.