Elad Gil on AI and how to discern real hype cycles from fake ones

So I think Silicon Valley has always gone through real hype cycles and fake ones. And by real, I mean it translates into something that actually has enormous impact. And fake ones are things where everybody thinks it's a thing and then nothing happens. And so I'll give you some examples, like between 2010 and, say, 2015 or something, you had a hype cycle around social, which turned out to be real, because you still had a bunch of companies founded right around then. You had Pinterest and you had Instagram. Spotify also. Spotify had a bunch of stuff, right, that was really interesting. Social products. That was a hype cycle. Mobile was a hype cycle. And you had Uber and DoorDash and a bunch of other companies. SaaS was a hype cycle and it turned out to be a real thing. And then you had a bunch of fake hype cycles, right? You had EdTech, which has the same problem of healthcare, which nobody will pay for it in the US at least. You had iot. You had Geo, like everything was going to be a geolocation app. So there's all these hype cycles that were fake ones and they were overlapping with real ones, and it was a lot of them in a short period of time, right? And so the hard part sometimes is to ask what's real and what's fake in terms of these hype cycles, because the real hype cycles actually lead to giant outsized companies or big changes in behavior or in user interface and adoption and usage and all sorts of things and companies that matter. And then there's fake ones. And so I think the AI wave is a very real one, and I actually think it's going to get much bigger in the next year as a hype wave. I think there's two reasons for that reason. One is you see companies getting massive velocity in terms of either revenue or usage from basically zero about a year ago. That's MidJourney. That's ChatGPT MidJourney launched in what, June, July of last year. ChatGPT launched in November. Character, brand new company. There's all these things that are getting enormous traction, is starting to get some good traction. So you're like, wow, these things are actually getting real usage and they're real products and they're real products, right? And some of them may be sustainable, some may not, but it's real. People care. And so, you know, it's a real thing because you see real usage, right? It's not just people writing articles. The other piece of it is so little time has passed since the seminal events in this industry. And I think the place where people get, where people mix things is you go back. So I've been investing in AI and I worked on AI products at Google, I worked on AI products at Twitter. Like the search team worked for me. All this stuff, I invested in ML companies, we called it ML, right, For a decade. And nothing worked for startups. All the value went to incumbents. And that was in the CNN and RNN world. And it just wasn't. The technology wasn't good enough to overcome the capabilities of incumbents. You just couldn't build stuff that was differentiated enough for the AI stuff to matter. And then we had two big breakthroughs, right? We had diffusion models in 2015 and then we had the Transformer architecture in 2017 and it took a couple years to scale those and do all the things that you need to do to bring technology to market. And then GPT3 came out two years ago, two and a half years ago, years ago. And then the starting gun was chatgpt, I think when people really woke up, right? Obviously last year, November Stable Diffusion launched in the summer or whatever, right. And so like it's been since November, which is eight months since ChatGPT came out and GPT4 came out in March, right. Four months ago, five months ago. A planning cycle for an enterprise. If you're a big business planning to use AI, your planning cycle is six months, right? And so the starting gun was eight months ago. You kind of dithered on it. Is this interesting enough? Maybe you start talking about it this year and then you're planning for six months and then you're going to prototype for a year and then you're going to actually launch something. So all the big enterprises haven't done anything yet and it makes a lot of sense for them to do nothing yet. And so that means tooling companies are going to pick up as these things pick up. And it means different applications will emerge with enormous uses that nobody's thought about. And mean, it's just so early. And so I just think we're going to see this run of hype as we see usage run, as we see enterprise adoptions, and it's going to take years. But I actually think we're very, very early in the hype cycle and I do think we'll have some trough of despair at some point. But you're not convinced it's going to be. I don't think it's anytime soon because I think it's just starting, yeah.