Evidence receipt / evaluation
Published · transcript-backedAlessio Fanelli: evaluation
15 Nov 2024 Latent Space Agents @ Work: Lindy.ai
“When I tried to set up Slack, it was like, hey, give me access to all channels and everything, which for the average person probably makes sense because you don't want to re-prompt them every time you add new channels.”
Source trail
Everything needed to verify it.
- Speaker
- Alessio Fanelli
- Attribution
- Verified speaker
- Claim type
- evaluation
- Recorded
- 15 Nov 2024
- Publisher
- Latent Space
- Episode
- Agents @ Work: Lindy.ai
Transcript context
…I actually used to dislike when people said, oh, text is not a great interface. I was like, ah, this is such a mid-take, I think text is awesome. And I've actually come around, I actually sort of agree now that text is really not great. I think for people like you and me, because we sort of have a mental model, okay, when I type a prompt into this text box, this is what it's going to do, it's going to map it to this kind of data structure under the hood and so forth. I guess it's a little bit blackmailing towards humans. You jump on these calls with humans and you're like, here's a text box, this is going to set up an agent for you, do it. And then they type words like, I want you to help me put order in my inbox. Oh, actually, this is a good one. This is actually a good one. What's a bad one? I would say 60 or 70% of the prompts that people type don't mean anything. Me as a human, as AGI, I don't understand what they mean. I don't know what they mean. It is actually, I think whenever you can have a GUI, it is better than to have just a pure text interface. And then how do you decide how much to expose? So even with the tools, you have Slack, you have Google Calendar, you have Gmail. Should people by default just turn over access to everything and then you help them figure out what to use? I think that's the question. When I tried to set up Slack, it was like, hey, give me access to all channels and everything, which for the average person probably makes sense because you don't want to re-prompt them every time you add new channels. But at the same time, for maybe the more sophisticated enterprise use cases, people are like, hey, I want to really limit what you have access to. How do you kind of thread that balance? The general philosophy is we ask for the least amount of permissions needed at any given moment. I don't think Slack, I could be mistaken, but I don't think Slack lets you request permissions for just one channel. But for example, for Google, obviously there are hundreds of scopes that you could require for Google. There's a lot of scopes. And sometimes it's actually painful to set up your Lindy because you're going to have to ask Google and add scopes five or six times. We've had sessions like this. But that's what we do because, for example, the Lindy email drafter, she's going to ask you for your authorization once for, I need to be able to read your email so I can draft a reply, and then another time for I need to be able to write a draft for them. We just try to do it very incrementally like that.…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.