This transcript is generated with the help of AI and is lightly edited for clarity.
REID:
I am Reid Hoffman.
ARIA:
And I’m Aria Finger.
REID:
We want to know what happens if, in the future, everything breaks humanity’s way.
ARIA:
Typically, we ask our guests for their outlook on the best possible future, but now every other week, I get to ask Reid for his take.
REID:
This is Possible.
ARIA:
Reid, no surprise, we are going to start off today talking about GPT-5. Our coworker Parth, I don’t think slept for the five days leading up to it. Everyone was so excited to see what’s to come, and this was a different release than the previous. People who watched the rollout weren’t sort of wowed as much by the technological changes. But what happened is this was an all-in-one. They deprecated all the previous models that people had come to know and love, but also be made fun of for whether you should be using 3.0, or -4, or the one that thinks, or the one that does all these other things. Which I think sort of boggled the minds of some more regular people who are trying to figure out which AI to use. What is just your first take on the GPT-5 launch?
REID:
So this might be a paradox, but I was expecting both a little bit of the blowback, and I think the GPT-5 launch is a very good thing. Ultimately, it’s almost akin to—I wonder how many people will remember this—when Facebook launched Newsfeed, there were actually protests outside of Facebook’s headquarters.
ARIA:
I mean, people went truly insane. For people who don’t remember, it was crazy.
REID:
And I was like, “Nope, Newsfeed’s a very good idea. It’s a good thing. It will continue.” Very quickly, everyone went, “Oh, I really like this.” And so that’s a little bit of the parallel in terms of what I think about, like GPT-5. I think it’s the, when you make these changes where people are familiar with the previous thing, there’s a bunch of people who are unhappy about the new thing and want to stay at the old thing. Whether it’s a change in interface on your computer or phone, or anything else. And it’s like, “Ah, that’s just not better because it’s different.” And there are a whole bunch of GPT-5 increases. I think Ethan Mollick told me the one that I think is the most relevant broadly, which is, there’s more already intelligent model direction.
REID:
And since most people don’t actually know how to choose models well for which kind of problem and thing they’re trying to do, the fact that you have a little bit of AI intelligence on, “Oh, this one you should use the more difficult thinking model for doing it.” And it doesn’t do it all on everything. For example, still with GPT-5, I find myself going and clicking the deep research button to make sure it’s doing that. And a little bit of the fun hack is to put “think deeply” in your prompt as a way of doing it. Because obviously, to some degree, the intelligence algorithm is also driven by cost for OpenAI. And so it’s going to try to get you a good result, but it’s going to default to the inferentially cheaper model. I think it is a substantive increase. I haven’t yet seen where the real question on coding and healthcare is, what is the network of thousands to tens of thousands of people in doing it in comparison.
REID:
Like, I haven’t yet seen the rollup of that. I know that OpenAI believes that it’s trained the current best, and I think a lot of people are like, “That’s just really good.” Now, of course the other part of the GPT-5 launch was the fact that OpenAI said, “Hey, we’ve got our current best models that are there through the service. And then we also made some open source models just to show you that the open source models we can make ones that are better than anything that’s currently out there, and enable them, and put them out there,” I think has also been a very significantly noteworthy thing.
ARIA:
I think one thing that’s interesting is there’s two different directions. One is you want to get the best possible models out there, people on the cutting edge using it. And then you also want really broad adoption. You want people to have good models when they’re using the free version. You don’t want them to have to think, to your point, about, “Ah, do I click deep research? Do I not?” OpenAI, honestly, with this launch, seems to be winning on both counts. Why is OpenAI ahead of the other frontier labs on this one?
REID:
Well, the key thing that OpenAI started very early, and they built on some work that was done at Google. This was part of—Ilya [Sutskever] and Greg [Brockman], and Dario [Amodei] and OpenAI went, “We are going fully at scale. And we’re going to think it’s just scale compute, with various patterns of learning, and scale data, and we’re going to apply whatever scale team we need to, and we’re going to aggregate that, and we’re going to go for that.” And they started that earlier than even Google, which had created the baseline techniques but was not as convinced about scale, was not convinced about productization, and so was not as fully into it—even the initial side. And then we’re followed by Microsoft in the combination of doing the deal with OpenAI, and then learning and doing that. And so they have the scale path, and the learning to—how to do the scale path as the advantage, and they continue to blitzscale along.
REID:
I mean, it is—in a sense—a classic Silicon Valley blitzscaling story, which is, bet on the fact that the scale version of this is going to really work. So invest in ways that people think you’re crazy. Go after scale team. Like “Yep, we’re going to move from 50 people to 500 people really, really fast. We’re going to put all that together. We’re going to put that there.” Do scale data. Which is like, “We’re going to take data from wherever we can legitimately get it—everywhere on the internet, Common Crawl, everything else. We’re going to do whatever deals we can, and we’re going to be throwing as much data at this. And we’re just going to continue to blitz on all of these rungs.” And what’s more, even though the ChatGPT rollout was a research rollout was not a launch, once they saw it, they would blitz on the ChatGPT deployment as well. And that’s part of the reason why OpenAI maintains its lead, and part of, in a baseline OpenAI strategy is to say, “We’re ahead on this scale, compute the deployment, et cetera—let’s just keep doing that.
ARIA:
Well, I think one thing that’s interesting is ChatGPT seems like such an overnight success. Like you said, they did the research launch, and then turned into an actual launch. Such like a classic Silicon Valley hockey stick. And I’d say another company that has been in the news recently that has that also classic, nowhere, nowhere, nowhere, nowhere, and then everywhere, is Figma. They famously took nearly a decade to go from their MVP to somewhat mass adoption. And for those of you who don’t know, it’s the design tool of choice for startups and large enterprises. And they were famously blocked by Adobe’s attempted acquisition, and they just went public recently—one of the biggest one-day gains in stock market history. And so they’re one of those companies that took over ten years to get to a real revenue place. And yet with AI, we’re seeing companies go to a hundred million ARR in a year or two. Are the Figmas of the world—the slow growth until you get mass adoption—are they the thing of the past? Do you have to have revenue immediately in these AI days?
REID:
Well, I think you certainly don’t need to have revenue immediately. I mean that’s part of what venture capital and capital market’s about, part of what Silicon Valley has understood in venture capital well before all the rest of the world. Like, actually in fact proving your strategic value in various ways, and then potentially getting the revenue. Now you always have a revenue thought, and you have to get there for long term, but like getting there strategically first. People don’t remember this was early criticisms of Facebook, like, “Oh, it’s great for college students, but I’ll never make any money.” Four years of commentary on Facebook, as an example. And it’s multiple through these things. And it’s actually understanding that getting to strategic position first as your leading edge, where revenue may be trailing. And look, I have a soft spot in my heart—not just for Figma, which is awesome, and Dylan and the whole team are great—but also like this whole, “Oh, it’s nothing, it’s nothing, it’s nothing. Oh my God, it’s everything.” To some degree, this is a Greylock VC special. Because there’s not only Figma, there’s LinkedIn—which is very similar, which is dead relevant. “Oh, amazing.” There’s Roblox, “Oh, trivia, little kids thing. Trivia, little kids thing. Ooh! Major platform!” We tend to do this a lot at Greylock, and that pattern will continue. Now, you want to get to strategic value as soon as possible. And one of the problems with slow bakes is if someone gets ahead of you on a slow bake, then you could suddenly die out in the wilderness and never get to your hockey stick. Now those of us who have been close to Figma—inclusive of myself, inclusive of John Lilly, inclusive of a number of other people—we knew Figma was awesome, and was coming.
REID:
And it’s partially because it’s a product-focused—like, more and more of these things are product-focused entrepreneur. And Dylan was an intern at LinkedIn. We’ve known Dylan a long time. And the notion of that design is the front-end to how you work, generally, across a number of different things, is part of it. It’s not just like people say, “Oh, just like Adobe.” And it’s like, “Well, actually, there’s obviously some questions about where you’d use Adobe, and some questions of where you use Figma, but it’s also a design approach to collaborative work.” And that’s the thing that makes Figma interesting in this. And Dylan’s one of those very product founder, true north, choose the thing you’re doing, “How does it make happen?” And so we just couldn’t be more delighted for Figma having its first moment on the public stage with its IPO.
ARIA:
I mean, Dylan seems truly like one of those founders who’s not chasing the latest hot thing, but super smartly integrating AI as appropriate—making sure he’s true to this core of designers that they’ve always served.
REID:
By the way, one other funny thing to say about Dylan—just because your questions are reminding me of how much I like Dylan—when I did the Bitcoin Rap Battle, Dylan was one of the people we had in the audience.
ARIA:
Oh, that’s awesome.
REID:
As part of one of the people doing the reactions to Satoshi versus Hamilton.
ARIA:
We’ll make sure to include the link to the Bitcoin Rap Battle in the show notes because it’s amazing, hilarious, and prescient—because it was many years ago. So speaking of another hot company, Perplexity has this eye-popping rise in valuations at least, from 520 million to over 18 billion in just over a year. And they also just made headlines with a $34.5 billion offer to acquire Chrome. A lot of people think this is just a marketing stunt—they want to grab attention. What is your read on this offer to acquire Chrome? Is it real, or is it just marketing?
REID:
Well, if Google turned around and said, “Sure,” I think Perplexity would do it. So it’s real in that regard. I think it’s not real in the regard that—completely speculating as an outsider—I think that maybe Perplexity might have reached out to Google beforehand—before it made a public offer—or might not have. Who knows? And I’m pretty sure Google’s not interested in this for a whole wide variety of reasons. And I think that given Perplexity is one of the efforts that’s trying to compete with Google, on “What does AI mean for search?”, I think it would be very low on Google’s list.
ARIA:
And so how does a company like Perplexity, which certainly has users, and has name recognition, and people are talking about it, but they’re going up against the behemoths. How do they differentiate? Is it specialization in the longterm? What’s the long play for them?
REID:
I’m not close enough to Perplexity to have a specific answer for Perplexity. It’s pretty amazing what they’ve accomplished. It’s, “Take risks that the incumbents won’t take.” It’s, “Move fast on various things.” It’s, “Try different distribution strategies.” It’s, “Do product direction,” that, for example, the incumbents can’t really do because they’re also servicing their current customers as a ways of doing it. It’s the classic Christensen Innovator’s Dilemma, which is, initially, the things seem small, or less relevant, but you bet right, and it grows large. I think all of those things would be naturally things to consider within the Perplexity set. I did think, by the way, the marketing stunt was clever. No, it’s a clever move that also comes at Google’s expense. So it’s not the kind of thing that Google’s got to be particularly happy about as a stunt. But you know, all fair.
ARIA:
Right? Typical incumbent versus entrant—figure out what you can do. So I want to end talking about one of the true OGs in the space. I remember very clearly, I think I was in eighth grade, and I made an “all starz”—with a Z—23 IM chatname. And it was all the rage among me and my eighth-grade friends. And just recently, AOL finally said no more to the dial-up connection that we all know so well. Those of us of a certain age can probably even sing the dial-up song that we were inundated with when we were young. And so what does this mean for the cycles of technology? It was used for 30 years. Some people might say, “Of course, Google, Microsoft—these tech giants—will persist 30 years into the future.” Do you think these cycles are getting shorter? Or no, we’re still going to see these technology companies being used 30 years into the future just like AOL was.
REID:
Well, one thing you’re entertaining me with, with your particular eighth-grade self, is back then the wisdom was anyone who claimed they were a teenage girl was actually a 30-year-old man.
ARIA:
Fair. Fair.
REID:
So you were actually one of the actual teenage girls using it.
ARIA:
That’s correct. That’s correct.
REID:
I think the short thing is that people over-predict, in the new things, the death of the old. Like, a classic one is when mobile started growing, people said PCs are over. And what happens is PCs grow—like mobile grows a lot more, but like PCs have continued. And actually, in fact, if you ask me, “Would they be continuing ten years from now,” the answer is yes, I think they will be. Because when it fits something that’s really useful, it actually continues, it has incumbents, it has a bunch of integrations, not just in enterprise but in people’s lives, and all the rest as ways of doing this. I myself, as you know, use—of course, I’m old—PCs, and laptops, and all the rest, more than I use mobile.
REID:
I actually deliberately use mobile somewhat just to be familiar with the experience, and to know where that’s going. But I think as people begin to use these things for more intensive activities. For example, how many people write their PhD on their mobile device? And it’s like not that many. Obviously, the mobile device goes with you everywhere, and you go, “Oh, I can respond to this email right now,” and all the rest like that. That’s part of the whole thing. So that’s part of the growth you see. And this is I think a very standard part of people’s misunderstanding in technology. For example, one of the memes right now is “vibe coding” is going to wipe out productivity software. And it’s like, well, actually, what I think you’ll see is productivity software will continue, and then “vibe coding” is going to add on to it, and add a bunch—either a lot, or a little.
REID:
But it’s not going to be like suddenly productivity software is going to go away. And so that’s the pattern that people need to understand. Now, partially because we venture capitalists, and we public markets, when you look at all the valuation and everything else, you’re betting on the ten-plus-year future in terms of how this stuff plays. And so that’s why you go, “Well, if it’s going to grow massively, I want to bet on the things that are growing.” So I want to bet on mobile, or I want to bet on, maybe, “vibe coding”. And that’s the way that this plays out. But it’s a very standard pattern that what happens is it persists on. And then by the way, when it dies, it dies very quickly, in a smaller number of years. And it’s mostly because the organization no longer has the economic throw weight to keep it going. It’s like, “Okay, this class of mainframes, there’s just not enough revenue”. And then that mainframe goes away. I’m sure there’s probably still some IBM mainframes around somewhere or other. But that’s the way the death happens. And it’s like, “Okay, the economics around dial-up just don’t work anymore.” So there’ll still be some smaller ISPs probably doing this, in various regions, but generally speaking, everyone’s like, “Ah, why use dial-up? Let’s just try to do other connectivity solutions.”
ARIA:
And to your point, so often technology just makes the market bigger. It’s like people have been talking about the death of the book forever. And it’s like, “No, no, no, the book is still with us. We just also have eBooks, and audiobooks, and websites, and mobile sites, and all the rest of it.” And so we will pour one out for AOL, but it’s certainly not the death of the big technology company. Reid, thank you so much. Always a pleasure.
REID:
Yep. Always look forward to the next.
REID:
Possible is produced by Wonder Media Network. It’s hosted by Aria Finger and me, Reid Hoffman. Our showrunner is Shaun Young. Possible is produced by Katie Sanders, Edie Allard, Thanasi Dilos, Sara Schleede, Vanessa Handy, Alyia Yates, Paloma Moreno Jimenez, and Melia Agudelo. Jenny Kaplan is our executive producer and editor.
ARIA:
Special thanks to Surya Yalamanchili, Saida Sapieva, Ian Alas, Greg Beato, Parth Patil, and Ben Relles.