Scaled Up Transformer Models

Cohen: I expect a full spectrum co-am system to look, which is of course, to be clear, is still completely hypothetical, not such a system. But in general, this is the resulting system, I expect that has these properties and that it reasons like a human. And then importantly, it also fails like ahuman. It is bounded so you can know what it won't do ahead of time. These stories have to be causal. Like currently, you can ask GPT. Why did you do that? And it'll give you some story. Well, they're not strictly necessary, but they're the easiest way to get it done. They might just sound convincing

Transcript

Play full episode

Transcript

Episode notes

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app