OpenAI’s $7 trillion plan, Apple’s Vision Pro, and AI traps

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

BOB SAFIAN: Hi everyone, Bob Safian here. Every few months I connect with Reid Hoffman to get his perspective on the key inflection points in the business world. Our most recent conversation, which we’re sharing today, starts with OpenAI’s new text-to-video tool SORA, and why OpenAI CEO Sam Altman is trying to raise $7 trillion. That’s trillion with a T.

We also talk about Meta’s dramatic resurgence, the Apple Vision Pro, three things driving tech layoffs, the AI trap that many businesses are falling into and more. As always, Reid’s insights both anchor me and get my wheels turning. I hope they do the same for you. Let’s get to it. 

THEME MUSIC

SAFIAN: All right. We ready for this?

HOFFMAN: Always with you, Bob.

Copy LinkThe scale lesson behind OpenAI’s Sora

SAFIAN: This past week, the new text and video tool that OpenAI released, Sora, which I don’t know if you’ve played with it all yet, I played with it a little bit and it does some amazing stuff.

HOFFMAN: What people have been working on pretty intensely for the last year or two is these multimodal generative models. And it doesn’t surprise me that OpenAI is the first to release the amazing new jump in capability. 

SAFIAN: They just keep seeming to be ahead of everyone else, even though so many entities are racing to try to leapfrog them.

HOFFMAN: Well, I think one of the key things that OpenAI has stayed true to, that relatively few of the other players are internalized to the degree to which OpenAI does it, which is a scale compute. So frequently, of course, people want to talk about the narrative of what’s going on in AI, as we’ve invented the new algorithm. 

And there are new discoveries and new algorithms, but really what we are is applying the scale compute lesson. You’ve got these news stories about Sam Altman going around for the trillion dollar chip fund and all the rest because he and the organization has embraced the “we are playing to scale” and by having played to scale, and there’s a lot of kind of interesting academic results and other things that say, hey, scale is kind of trumping all yes, this algorithm is a little better. Yes, this algorithm is a little worse. Yes, this data set is a little better. Yes, this data set’s a little worse, but it’s the scale that is primarily driving. And so it drove it in the large language models, hence large, out of language models, and it’s what’s driving it in the multi modal as well. 

SAFIAN: Once though you have that scale compute, it allows you to create these executions in a way that if you don’t have that scale, it’s not that you can’t, but it’s not going to be smooth the same way, it’s just going to be that much harder or that much bumpier getting to those places.

HOFFMAN: Yes, exactly. The Sora work is amazing. I don’t want to take anything away from the fact that it was smart, high-quality people working intensely in a team.

But it’s also: creating these super expensive computers that’s a bold risk taking move from both OpenAI and from Microsoft.

Copy LinkReid Hoffman on OpenAI’s attempt to raise money for the chip industry

SAFIAN: And now you mentioned Sam Altman and his, I think it’s 7 trillion he’s trying to raise for the chip industry. Which is like a wild number. Right? 7 trillion dollars, especially at the same time that he is running this company. Is it about that, we really need that number to get to the future of, you know, this industry, or that’s a target that pushes us in the right direction?

HOFFMAN: What I would say is they’ve probably done some internal calculations about what scale gets them potentially to artificial general intelligence. There’s several different definitions of artificial intelligence, but obviously the rough thought is something intelligent, like we are, capable of having  metacognition and strategic planning and one shot learning and a bunch of other stuff.

And that the scale approach of this will get there. So my guess is that’s how the number got calculated just from knowing the people. Now, that being said, part of the thing I think is amazing is whether or not you think AGI is high probability, medium probability, low probability, wherever you are on that spectrum, you will be creating really interesting increases in cognitive capability of these systems.

Say they can’t raise seven trillion or not in one bold swoop or whatever else, you’re still going to have amazing progress. So kind of doesn’t matter if it was set out as we need this to get to AGI or it’s a target number, with aspirational focus, that continued investment to scale is going to be one of the things that’s going to net a lot of the steam engine of the mind revolution and this amazing cognitive amplifiers for human work that will come out of some portion of that, whatever portion of the goal plays out.

SAFIAN: Sometimes I feel like the story in technology development is like it’s software, which feels like it’s been more recently, but now with AI, of course, it’s software, but it sounds like it’s the hardware also. These two things are being married or relying on each other to advance in a way that maybe we haven’t had for a little while. 

HOFFMAN: Well, I think it’s always been a combination of software and hardware. We were operating for decades on so called Moore’s law, really kind of Moore’s hypothesis or principle or something, which by doubling the number of transistors and getting the kind of compute better, we could continue that software progression.

And there’s some really compelling graphics that kind of show that the progress in this modern way of AI came about from when, as Moore’s law flattened, it just changed its shape of what the hardware loop was. Was opposed to the hardware loop being we’re also doing Moore’s law. It’s the – we’re doing massive parallel configurations. By the way, what that means is, that tends to go to certain kinds of algorithms.

It requires a learning versus a programming, artifact tends to be probabilistic computing. There’s a stack of things that kind of go into how that scale plays. But that the hardware software combination continues.

SAFIAN: Sometimes sort of the valuations rise for one group and then the other group, right? They sort of seem to sometimes go in sequence. Although right now with Microsoft on the one hand and OpenAI and with Nvidia on the other end, they’re both going.

HOFFMAN: Valuations, are actually, in fact, kind of a market prediction about where future value will lie and you go, okay, the hardware, like NVIDIA has got these great business with 80 percent margins, we have this fairly strong hardware edge that so far we have not had anyone catch us.

And so people go, okay, there’s a lot of value there. We’ll bet on that. But on the other hand, people also know there’s going to be a ton of value on the software side. Right. And so the short answer is the market saying we’re betting on both the software and hardware on the AI side.

And so place your bets.

Copy LinkIs the AI industry creating a David and Goliath scenario?

SAFIAN: And who wins in the long run or who is the bigger winner? It’s kind of hard to tell right now. I mean, right now we see OpenAI obviously winning, but, and I’m thinking of the earlier waves of social media, like Facebook was not necessarily considered to be the winner in the beginning.

And yet they ended up being the winner. This happens in business and in technology in particular.

HOFFMAN: 100%. And also it’s a little bit reminiscent. Also, the question is, is this going to be on a technological change that large companies win from or start-ups win from? And the answer that I’ve been given fairly consistently along this whole path is both, right? It isn’t going to be the David and Goliath where David invents this new thing AI and it kind of resets the apple carts of the Goliaths that are appropriately bought into AI, which, is intensely Microsoft and Google and some Amazon, definitely Facebook. 

The AI will be a massive increase for the large companies, but will also be very valuable across a whole set of startups. And so I think it’s going to be a realization of the software revolution, that transformation of industries by software across the entire stack of size.

And I think that’s also true hardware and software.

SAFIAN: The concentration of wealth right now in the top tech names: Microsoft, NVIDIA, Apple, Amazon, it’s like 25 percent of the market value for the S&P 500. I mean, it’s not been that way before. Is that troubling to you?

HOFFMAN: Well, there’s definitely places where scale can be not good. That’s an odd statement to be making on the Masters of Scale podcast, but it’s not what most people think. Most people think – oh, big is bad, oh, look it’s continuing market dominance of the top tech companies.

And you’re like, well, actually, in fact, if we were five big U.S. tech companies, or seven big U.S. tech companies, heading to three. I’d be actually quite concerned. We’re actually five or to seven heading to 10 to 12. And the competition between these organizations is fierce, and it creates a lot of opportunities for start-up. It creates a lot of services for consumers. It creates a lot of value within the American tech industry, which benefits America. 

Scale is concerning when it distorts it because it crushes competition. If tech company X, where they’re going to get to scale and then lock out other players to the detriment of society, detriment of industries, detriment of consumers but that’s not an absolute scale number. That’s a relative scale number to your competitors and other players. That’s the mistake that lots of press and everyone else just mistakes. 

That’s the reason why if it was five to seven going to three, well, shit, if it’s five to seven going to 10 to 12, then it’s good. NVIDIA’s inclusion in the trillion dollar club is part of the going from five to seven to 10 to 12, right? That’s the instance of it, like I’ve been saying this for years and look here’s an example proof. And he’s like, well, you didn’t say NVIDIA before. It’s like, you don’t know which ones, but you know that the competitive ground swell is coming.

And by the way, one of the benefits that you have when you have this is all of these companies are investing massively in technological R and D.

It’s not what people frequently are saying. Scale is bad. It’s saying you have to watch for certain elements of scale, which I don’t think we are actually triggering yet.

SAFIAN: So this is like qualitatively different than say, regulators looking at the airline industry and saying JetBlue and Spirit shouldn’t get together, or the grocery industry, Kroger’s and Albertson shouldn’t get together, the tech industry is sort of qualitatively different. 

HOFFMAN: Well, if you were saying, Microsoft and Google should combine into one company. I would definitely agree that would not be a good idea, right? But if you said, should, large company X be able to buy small company Y the answer is sure.

They should be able to, the competition between these organizations is ferocious. And by the way, as a venture capitalist, my primary identity, there’s a lot of opportunity on the startup side.

Back in the day, decades ago when Microsoft was the one big tech company, that was a bigger challenge. It was good to limit it. But I think everyone was surprised by how quickly through the browser and through other things that new tech companies could emerge and challenge it.

Now, the last point to your earlier question, these top tech companies are a quarter of the S and P.

I think this is beginning to really show the realization that all companies are heading towards becoming tech companies that you need that tech amplification in all of it. And that’s what we need to be figuring out across all industries. 

And it’s the thing that we need to be embracing for the future of our economies, the future of our prosperity as societies, individuals, the classic dialogue thing is we got to limit the big tech.

And you’re like, well, that’s if you want to limit your economic future. Right? It’s like, no, no, what you want to do is say, how do we leverage the fact that we have some technological advantages to benefit society broadly?

And that’s where the intellectual work needs to be.

SAFIAN: Reid is such an empassioned advocate of technology, but that doesn’t make him wrong. The future is coming faster than ever, and you want to be on the right side of it. After the break, we talk about Mark Zuckerberg and Meta, the Apple Vision pro, tech layoffs and more. Stick around.

[AD BREAK]

SAFIAN: Before the break, Reid Hoffman shared his perspective on how AI is altering the game of scale. Now we talk about Mark Zuckerberg and Meta, the Apple Vision Pro, tech layoffs, and the trap that many business leaders are falling into in 2024. Let’s jump back in.

Copy LinkReid Hoffman on Mark Zuckerberg’s adoption of AI 

So I want to ask you about Mark Zuckerberg and Meta. Facebook was the quintessential social media company. And then it sort of seemed like they were ceding ground.

Particularly the TikTok, right? Which sort of captured the cultural heat. Zuckerberg’s talking about the metaverse since spending a lot of money, maybe not getting a lot for it. The Apple Vision Pro is getting all the buzz more than Quest VR did and Sheryl Sandberg, steps aside as COO and then she’s leaving the board.

And then on one day, Meta stock bam, goes up like 20%. What happened? What did people sort of misappreciate or misunderstand? And I don’t know whether it’s about Meta or about the industry or about Mark. 

HOFFMAN: It didn’t surprise me, although I’m not a public market trader so I tend to be structuralist over years, what is the position over, one year, three years, five years, 10 years, it’s part of why doing early stage venture. Now, one, Mark is a bold innovator, will take bold bets.

Some of the bold bets are pretty amazing. Like when he bought Instagram when it was really small, bought WhatsApp, some of the bold bets, I personally don’t think work out as well, Oculus and Meta is their focus, but he’s a constant infinite learner. And so I saw when the light bulbs came on of “oh my God, this AI thing is going to be really important and we should be doubling down on that”. I don’t think they’re agents and so forth are that good yet.

But the notion of how do you tie the AI stuff into the advertising system and have an alternative, really compelling advertising system, how do you continue to do engagement with the feed and other kinds of things, which I think they do now.

Like, it didn’t surprise me at all that they would have a. It’s a very, very vigorous recovery, uh, given Mark, given a focus in AI, given a natural set of, of assets in aspects that are important to human life.

SAFIAN: I mean, he made the business more efficient, I think than people expected, and I think the stickiness of the different parts of Meta proved to be stronger than certainly the marketplace was anticipating. I mentioned the Quest and Apple Vision Pro, have you tried these things out? I know Microsoft has its HoloLens. Is this an area that you play with that you dabble in, or is it not really your jam?

HOFFMAN: Well, my very first product management job was in virtual worlds. Having gone into the promise of it and realized it was so underdelivered that I’m a little bit overly skeptical. I’m part of the reason why Greylock didn’t invest in Magic Leap and other things because I go, look, there’s really amazing technology, but I just don’t see it coming together as the new tech platform yet.

Now, I’ve had a couple of my trusted friends, David Z, other people say I really need to play with Vision Pro, I need to see what it is, I have played with the Oculus, I have played with HoloLens, I have yet to think that any of these things are a new platform. I definitely think that the economic cost of the Vision Pro means it certainly won’t be a platform yet.

Now, will it be, as some people are speculating, is it a limited end number of iterations from here to there? To make it work, is it two or three? And I think that’s part of the reason why David has been, like, okay, you really need to go play with this. And so I’ve ordered one to get a sense of it.

The cautionary thing about how you get to platform. Here we are, you and I talking, you know, wearing glasses.

SAFIAN: Yes.

HOFFMAN: Even these things, which are an amazing part of early technology, a number of human beings pay 5, 000 to have a laser applied to their eyes. So they don’t have to wear these things. So that’s what you have to clear in terms of the value proposition to get to a general platform, and if it isn’t a double digit percentage of humanity, then it’s not going to be a general platform. Now, it may be a work platform, maybe like a doctors, nurses, police people, fire people, other kinds of things is part of why I think the HoloLens and Microsoft kind of realized better than the consumer folks for these things but as a general platform, you have a very high hurdle to clear.

SAFIAN: The benefits really have to be powerful enough to make it worth the inconvenience of much as we love that we have these spectacles and they allow us to see they’re pain, they’re pain. 

HOFFMAN: Yes, exactly.

Copy LinkWhy are we seeing layoffs in the tech industry?

SAFIAN: In the optimistic, good times that the tech industry is having right now. There have also been a bunch of layoffs talked about at places like Amazon, and Meta, and Alphabet. And I’ve heard kind of three different theories about why these layoffs happen. One being, overhiring from the pandemic, which is something that Zuckerberg has said at Meta they sort of overhired some.

Another argument is copycat layoffs, basically doing it for performative reasons to appease public market investors to make it look like you’re being sort of sharper. 

And the third being that AI is making things more efficient and that, these businesses just don’t need people the same way.

Now, obviously, there are many more reasons why layoffs could happen. I’m curious whether you have a theory about why this is happening. 

HOFFMAN: So the first one is certainly the case, the overhiring from the pandemic and kind of reconfiguration and so forth, because people overly generalized from – oh, during the pandemic, this is the new normal. It’s like, well, no, it’s not going to be the new normal.

To some degree, it wasalmost like a collective mistake where everyone’s seeing everyone’s hiring, so we should be hiring too. 

Now the copycat layoffs is actually a touch more sophisticated as a question, which is, I don’t think that any of the leaderships of any of these companies is so banal as to simply be, well, because the investors are demanding it from margins and other people doing it, I have to do it too.

And so it doesn’t say there isn’t something there, but I think what actually in fact goes into this is when there is that zeitgeist, a time of layoffs, you can then kind of go, okay, I now have a permission to do some of the resetting in the business. Like I should be doing less of X, more of Y.

SAFIAN: Things I should be doing anyway, but suddenly I can do it and it’s sort of part of the zeitgeist and I’m not gonna pay as much of a price for it.

HOFFMAN: Exactly like, because the price you usually play is that you’re the only person doing these kind of layouts. Is it because you suck? You’ve made bad decisions! And so there’s a pressure to not do it in bold stroke. If we do that and we’re standing out there by ourselves, it causes a bunch of negative speculation. Whereas when we go ‘oh shoot,’ we can now take opportunity. Like the industry itself is doing layoffs ,and we can do some reconfiguration. And I think that’s much more of what you’re happening. 

It’s all kind of simplistic press stories and what you should really do is look underneath. Okay, what are they reshaping their businesses for? Now, they may be reshaping their businesses for an anticipated AI productivity. I think there will be a bunch of AI productivity.

I don’t think that much AI productivity is currently baked into the numbers and rigged in the way that people are operating. So it’s certainly not a post-fact thing. 

I think the other parts of the consideration are this gives us a way to reshuffle and that reshuffling that reconfiguration is what you should be looking at in each of these tech companies. 

SAFIAN: I do feel like after over the last four years between pandemic and supply chain, and inflation, and AI, that a lot of leaders that I’m talking to are sort of, they’re not saying that they’re resetting or they’re like, let’s not be too aggressive, but they’re sort of like reflecting a little bit right now and I don’t know, I wonder sometimes where that’s complacency or that’s fatigue or that’s wisdom?

HOFFMAN: Well, frequently with these choices, Bob, as you know, it’s a combination of all of them and it depends on the different leaders, right? To essentially go to what people should be, as part of the resetting is, should be saying, look, we are at another wave of technological transformation.

We’ve had a bunch, we’ve had the internet, we’ve had the iPhone, we’ve had mobile computing, we’ve had cloud, now, AI is going to take all of those to another level. The reason why it’s so big is it’s a compounder of all of them together. And that’s coming in a small n number of years.

So if you’re not plotting part of the technology strategy of your business to be doing this. I don’t mean IT strategy, you know, it’s the technology underlying how your whole business operates: what your supply chain is, how your your employees work together, how you sell and market, how you build and constitute your product, how you do innovations, how you do financial analysis – all of that is an evolution from essentially the steam engine of the mind and AI. And so the advice that I give everyone, they say, well, what should I do now? Look, it’s all in very dynamic flux.

You can’t just say it’s only X right now. And that’s all you need to do. No, so what you need to do is start experimenting with it. So the one thing about the reflective thing where it’s lazy is no, no, you should be diving into the experimentation, not necessarily committing fully to a particular path right now, because you don’t really know.

But if you’re not experimenting with some vigor, you’re likely to be making a pretty dangerous mistake.

SAFIAN: And if you’re sort of waiting for things to become, quote, clear, which is what a lot of leaders, I think, want to do, they’re actually losing ground because they’re not getting comfortable with how this new technology can change the way they operate.

HOFFMAN: Exactly. You should not be waiting. You should be experimenting, possibly experimenting vigorously.

SAFIAN: Well, Reid, thanks for doing this. Always good to chat with you.

HOFFMAN: Always great to talk with you, Bob. I look forward to the next. 

SAFIAN: What I came away with most from this discussion with Reid is the potential trap of waiting for the “right moment” to engage with a new opportunity or a new challenge. 

If we’re waiting for clarity to emerge in today’s fast-changing world, then we’re waiting too long. We have to jump in and experiment, in the face of ambiguity. I’m Bob Safian, thanks for listening.