Michael Royzen: belief

Transcript context

…What do you think is the most interesting unsolved question in AI? I think solving hallucinations, being able to guarantee that the answer will be correct is I think super interesting. And it's particularly relevant to us because like we operate in a space where like everything needs to be correct. Like the code, like not just the logic, but like the implementation, everything has to be completely correct. And there's a lot of very interesting work that's going on in this space. Some of it is approaching it from the angle of formal grammars. There's a very interesting paper that came out recently. I forget where it came out of, but the paper is basically you can define a grammar that restricts and modifies the model's log probs, like decoding strategy to only conform to that grammar. And that helps it... Is this LMQL? Because I feel like LMQL is a little bit too structured for... If the goal is avoiding hallucination, that's such a vague goal. Yeah.…

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