Jesse Silver: evaluation

Transcript context

…That makes sense. And what about the underlying models? What's the prototype on OpenAI? And then you went on a open source models, like how much can you get away with, with the commercial models? I know there's a lot of, you know, RLHF, have you played around with any of the uncensored models like the Dolphins and things like that? Yeah. Any insight there would be great. Yeah. Well, I think you can get reasonable outcomes on sort of the closed source models. They're not very cost effective because you may have very, very long conversations. And that's just part of the fan experience. And so at some point you need to move away if you're using OpenAI. And also OpenAI, you can almost like feel the OpenAI-ness of a generation and it won't do certain things for you. And you'll just continuously run into problems. We did start prototyping on OpenAI and then swiftly moved away. So we are open source. You know, in our workflow, we have modules that do different things. There's maybe a state machine element, which is if we're conversing, we're in a different state than if we're providing some sort of sexual experience. There's reasoning modules about the content to send. There's understanding the content itself. There's the modules that do the chatting. And then each of these relies on perhaps a different fine-tuned model. And then we have our eval framework for that. When you think about fine-tuned model, how do you build that data set, I guess? More like the data set itself, it's like, what are the product triggers that you use to say, okay, this is like we should optimize for this type of behavior. Is there any sort of analytics, so to speak, that you have in the product? And also like in terms of delivery, is the chat happening in the fan kind of like app? Is it happening on like an external chat system that the creator offers to the customer? And kind of like, how do you hook into that to get the data out? I guess it's like a broader question, but I think you get the sense.…

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