Mentions personal use of Claude. Mentions personal use of ChatGPT.

Transcript context

…Sure, the usual way I format things is I'll start with some data set of inputs and outputs. And it might be ratings for a pizza shop and some binary classification of like, is this a positive sentiment, is this a negative sentiment? And so this is going back more to classical NLP, but I'll structure my prompt as, Q, colon, and then I'll paste the review in, and then, A, colon, and I'll put the label. And I'll put a couple lines of those. And then on the final line I'll say, "Q, colon," and I'll input the one that I want to, the LLM to actually label, the one that it's never seen before. And Q and A stand for question and answer, and of course in this case, there are no questions that I'm asking it explicitly. I guess implicitly it's, is this a positive or negative review? But people still use Q and A even when there is no question-answer involved, just because the LLMs are so familiar with this formatting due to, I guess, all of the historical NLP using this. And so the LLMs are trained on that formatting as well. And you can combine that with XML. Yeah, there's a lot of things you can do there. That is super helpful. We'll link to this report, by the way, if people want to dive down the rabbit hole of all the prompting techniques and all the things you've learned. As an example, I use Claude and ChatGPT for coming up with title suggestions for these podcast episodes. And I give it examples of just examples of titles that have done well, and then it's 10 different examples, just bullet points. That's another thing you [inaudible 00:17:22]. You don't even necessarily have the inputs and the outputs. In your case, you just have, I guess, outputs that you're showing it from the past.…

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