Evidence receipt / evaluation
Published · transcript-backedJesse Silver: evaluation
16 May 2024 Latent Space Emulating Humans with NSFW Chatbots - with Jesse Silver
“I think we just return to, how do you generate interactions that fit a coherent role based on what the creator's brand is?”
Source trail
Everything needed to verify it.
- Speaker
- Jesse Silver
- Attribution
- Verified speaker
- Claim type
- evaluation
- Recorded
- 16 May 2024
- Publisher
- Latent Space
Transcript context
…Okay. That's an interesting discussion. We are going to ask a tech stack question straight up in a bit, but one thing I wanted to make sure we cover in this personality profiling question is, are there philosophies of personality? You know, I am a very casually interested person in psychology in general. Are there philosophies of personality profiling that you think work or something that's really popular and you found doesn't work? What's been useful in your reading or understanding? We don't necessarily use a common psychological framework for bucketing creators or fans into types and then using that to imply an interaction. I think we just return to, how do you generate interactions that fit a coherent role based on what the creator's brand is? And so there are many, many different kinds of categories. And if you just go on Pornhub and pull up a list of all the categories, some of those will reduce into a smaller number of categories. But with the diva type, you need to be able to prove yourself and sort of conquer this person and win them over. With a girl next door type, you need to be able to show yourself and, you know, find that they like what they see, have some relationship building. With a dominant type of creator and a submissive type of fan, the fan is going to want to prove themselves and like continuously lose. And so I think language models are good by default at playing roles. And we do have some psychological profiling or understanding, but we don't have an incredibly sophisticated like theory of mind element in our workflow other than, you know, reflection about what the fan is wanting and perhaps why the action that we took was unsuccessful or successful. I think the model that maybe I would talk about is that I was talking to a friend of mine about how they seduce men. And she's saying that, let's say she meets an older man in an art gallery, she's holding multiple hypotheses for why this person is there and what they want out of her and conversely how she can interact with them to be able to have the most power and leverage. And so are they wanting her to act naive and young? Are they wanting her to act like an equal? Why? And so I think that fans have a lot of alternatives when they're filtering themselves into fan platform profiles. And so most of the time, a fan will subscribe to 50 or 100 profiles. And so they're going to a given person to get a certain kind of experience most of the time. 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.…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.