High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / belief

Published · transcript-backed

Shawn Wang: belief

16 May 2024 Latent Space Emulating Humans with NSFW Chatbots - with Jesse Silver

“I think there's also a rise in interest in recording devices where you're effectively recording your entire day and summarizing them.”

— Shawn Wang

Source trail

Everything needed to verify it.

Speaker
Shawn Wang
Attribution
Verified speaker
Claim type
belief
Recorded
16 May 2024
Publisher
Latent Space

Transcript context

…Yeah. I think we have a lot of information about fan and so when a fan that's currently subscribed to one of the creators we work with, their profile subscribes to another one of the creators we work with profiles, we need to be able to manage sort of fan collisions between multiple profiles that a creator may have. And then we also know that fan's preferences, but we also need to ask about their preferences and develop our concept and memory of that fan. Awesome. Two more technical questions because I know people are going to kill me if I don't ask these things. So memory and DSPy. So it's just the memory stuff, like you have multi thousand turn conversations. I think there's also a rise in interest in recording devices where you're effectively recording your entire day and summarizing them. What has been influential to you and your thinking and just like, you know, what are the biggest wins for long conversations? So when we onboard onto a profile, the bar that we need to hit is that we need to seamlessly pick up a conversation with someone who spent 20K. And you can't always have the creator handle that person because in fact, the creator may have never handled that person in the first place. And the creator may be just letting go of their existing chatting team. So you need to be able to understand what the customer's preferences are, who they are, what they have bought. And then you also need to be able to play out similar sessions to what they might be used to. I mean, it is various iterations of like embedding and summarizing. I've seen people embed summaries, you know, embedding facts under different headers. I think retrieving that can be difficult when you want to sometimes guide the conversation somewhere else. So it needs to be additional heuristics. So you're talking to a fan about their engineering project, and perhaps the optimal response is not, oh, great, yeah, I remember you were talking about this rag project that you were working on. And maybe it's, that's boring, like, play with me instead.…

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

Search evidence