“In order to access this system, humans have to bid for time. They essentially have to say how much in these fake dollars they're willing to pay for their job to be run… And LLMs using this thing just really don't have a sense for how much their job is worth. And we've seen evidence of the LLMs thinking that their job is worth way more than it should be, and thus preempting a bunch of human work.”

…better access controls into various systems to make it so that the models can only do the things that we think are like actually safe. And increasingly, that also has us relying on, going back to the previous point, LLMs judging whether. or not something's safe, which is a very interesting thing.

…an access control system. But if you just like naively use LLMs and kind of set them up without thinking about the setup at all, then the LLMs just inherit all the permissions of the human. And that's super wrong. Like you don't want that. You want some kind of bounds on what the LLM can do. My

Yeah. I mean, I think the other thing that's important and hard to think about here is that to some degree, you should think about the models adversarially. Like they are not, you know, people talk about AI alignment and it's like a kind of big theoretical AI safety long-term kind of issue where in…

And that comes up in security. It also comes up in places like evaluations where we're like, you know, you're trying to judge the success of the model, but you have to remember, like you mentioned earlier, that the models are trained to get, you know, to do the thing that you're telling them to do.…

And this is when you see warnings, this is like compiler level, like unused variable warnings or whatever. And like warnings as errors means if you have a warning anywhere in your code, it just won't build. And we basically have a kind of firm-wide, you know, code-based wide practice that like…