Speaker unverified: belief

Transcript context

…We saw multiple demos in 2025 of, like, Claude can use your computer, or OpenAI had operator, and they all suck. So they’re investing money in this, and I think that’ll be a good example. Whereas actually, taking over the whole screen seems a lot harder than having an API that they can call in the back end. Some of that is you have to then set up a different environment for them all to work in. They’re not working on your MacBook; they are individually interfacing with Google and Amazon and Slack, and they handle all these things in a very different way than humans do. So some of this might be structural blockers. Also, specification-wise, I think the problem for arbitrary tasks is that you still have to specify what you want your LLM to do. What is the environment? How do you specify? You can say what the end goal is, but if it can’t solve the end goal—with LLMs, if you ask it for text, it can always clarify or do sub-steps. How do you put that information into a system that, let’s say, books a travel trip for you? You can say, “Well, you screwed up my credit card information,” but even to get it to that point, as a user, how do you guide the model before it can even attempt that? I think the interface is really hard. Yeah, it has to learn a lot about you specifically. And this goes to continual learning—about the general mistakes that are made throughout, and then mistakes that are made through you.…

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