Soumith Chintala: belief

Transcript context

…in my opinion. Whether it's transformative use. Yes. So yeah, I think this New York Times opening eye case is gonna go to the Supreme Court and we'll have to decide it because I think we never had to deal with it before. And then finally, for synthetic data, the thing that I'm personally exploring is solving this great stark paradigm difference between rag and fine tuning, where you can kind of create synthetic data off of your retrieved documents and then fine tune on that. That's kind of synthetic. All you need is variation or diversity of samples for you to fine tune on. And then you can fine tune new knowledge into your model. I don't know if you've seen that as a direction for synthetic data. I think you're basically trying to, what you're doing is you're saying, well, language, I know how to parametrize language to an extent. And I need to teach my model variations of this input data so that it's resilient or invariant to language uses of that data. Yeah, it doesn't overfit on the wrong source documents.…

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