Lenny Rachitsky: commitment

Transcript context

…Thanks, Lenny. It's great to be here. I'm super excited. I'm very excited because I think I'm going to learn a ton in this conversation. What I want to do with this chat is essentially give people very tangible and also just very up-to-date prompt engineering techniques that they can start putting into practice immediately. And the way I'm thinking about we break this conversation up is we do a basic techniques that just most people should know, and then talk about some advanced techniques that people that are already really good at this stuff may not know. And then I want to talk about prompt injection and red teaming, which I know is a big passion of yours, something you spend a lot of your time on. And let's start with just this question of, is prompt engineering a thing you need to spend your time on? There's a lot of people that, they're like, "Oh, AI is going to get really great and smart, and you don't need to actually learn these things. It'll just figure things out for you." There's also this bucket of people that I imagine you're in that are like, "No, it's only becoming more important." Reid Hoffman actually just tweeted this. Let me read this tweet that he shared yesterday that supports this case. He said, "There's this old myth that we only use 3 to 5% of our brains. It might actually be true for how much we're getting out of AI, given our prompting skills." So what's your take on this debate? Yeah, first of all, I think that's a great quote. And the ability to, it's called elicit certain performance improvements and behaviors from LLMs is a really big area of study. So he's absolutely right with that, but, yeah, from my perspective, prompt engineering is absolutely still here. I actually was at the AI Engineer World's Fair yesterday, and there was somebody, I think before me, giving a talk that prompt engineering is dead. And then my talk was next, and it was titled Prompt Engineering. And so I was like, "Oh, I got to be prepared for that." And my perspective, and this has been validated over and over again, is that people will always be saying, "It's dead," or "It's going to be dead with the next model version," but then it comes out and it's not. And we actually came up with a term for this, which is artificial social intelligence. I imagine you're familiar with the term social intelligence, describes how people communicate, interpersonal communication skills, all of that. We have recognized the need for a similar thing, but with communicating with AIs and understanding the best way to talk to them, understanding what their responses mean, and then how to adapt, I guess, your next prompts to that response. So over and over again, we have seen prompt engineering continue to be very important.…

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