High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / evaluation

Published · transcript-backed

Jesse Silver: evaluation

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

“And we also work with creators where tone is not important at all. And we find that a product like this is really good for this industry because LLMs are good at replicating tone, either handcrafting a prompt or doing some sort of K-shotting or doing some sort of fine tuning or doing some other sort of optimization.”

— Jesse Silver

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

…Two final questions. One, do the creators ever talk to their own bots to try them? Like do they give you feedback on, you know, I would have said this, I would have said this? Yeah. Is there any of that going on? Yes. I talk to creators all the time, every single day, like continuously. And during the course of this podcast, my phone's probably been blowing up. Creators care a lot about the product that is replicating their personal brand in one-to-one interactions. And so they're giving continuous feedback, which is amazing. It's like an amazing repetition cycle. We've been super lucky with the creators that we worked with. They're like super smart. They know what to do. They've built businesses. They know best about what's going to work with their audience on their profile. And a lot of creators we work with are not shy about giving feedback. And like we love feedback. And so we're very used to launching on a profile and getting, oh, this is wrong, this is wrong. How did you handle this person this way? Like this word you said was wrong. This was a weird response, like whatever. And then being able to have processes that sort of learn from that. And we also work with creators whose tone is very important to them. Like maybe they're famously witty or famously authentic. And we also work with creators where tone is not important at all. And we find that a product like this is really good for this industry because LLMs are good at replicating tone, either handcrafting a prompt or doing some sort of K-shotting or doing some sort of fine tuning or doing some other sort of optimization. We've been able to get to a point on tone where creators whose tone is their brand have said to me, like, I was texting my friend and I was thinking to myself how the bot could have said this. And transitioning from having a bad LLM product early on in the process to having a good LLM product and looking at the generations and being like, I can't tell if this was the creator or the product has been an immense joy. And that's been really fun. And yeah, just sort of continued thanks to our customers who are amazing at giving us feedback. Well, we have to thank you for being so open and generous with your time. And I know you're busy running a business, but also it's just really nice to get an insight. A lot of engineers are curious about this space and have never had access to someone like you. And for you to share your thoughts is really helpful. I was casting around for our closing questions, but actually, I'm just going to leave it open to you. Is there a question that we should have asked you, but we didn't?…

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

Search evidence