JORDAN McLEOD: In a world of tech and innovations race, where start-ups rise and some may fall, the masters of scale guide with grace, their lessons heard by all, their podcast a beacon of light illuminates the path to growth. With every episode, a new insight and wisdom for all to know.

BOB SAFIAN: Hey everyone. It’s Bob Safian here. And that voice you just heard was Masters of Scale executive producer Jordan McLeod reading part of a sonnet written by ChatGPT. And we’ll have more from ChatGPT later in the show. In fact, this entire episode is all about AI.

In these Need To Know episodes, Reid Hoffman and I have unfiltered conversations about the most important business topics impacting entrepreneurs right now. 

Normally we jump between a range of topics. But there’s no topic hotter right now than ChatGPT and AI. So for this installment, AI will be our focus. We’ll cover issues from the looming AI search wars, to how recent developments have upended the risk-reward analysis of embracing AI, and what leaders of businesses large and small should be doing right now to make the most of this AI inflection point.

Let’s get to it.

[THEME MUSIC]

SAFIAN: So, Reid, usually these Need to Know episodes bounce around a bit, touching on a bunch of different trends in the business world. But this time, we’re going to dig in deep on one topic because it’s so talked about right now: ChatGPT and AI.

You are an early supporter of OpenAI, which created ChatGPT. You’re also on the board of Microsoft, which is invested in and aligned with ChatGPT. You’ve recently started a new AI business, Inflection AI, with Mustafa Suleyman.

Reid Hoffman on ChatGPT — and its reception

But, I have to start with ChatGPT itself. Since it was introduced last year, it’s been a phenomenon. The number of users has ballooned beyond what anyone expected. It’s remaking assumptions in businesses, in education, in the media, in culture. This reception for ChatGPT, did it surprise you?

HOFFMAN: Not really. And part of that’s because I’ve seen what the magic is that’s coming with AI in large language models for a number of years. And part of what, over the last couple years, may seem like prescient, random predictions now coming true. It’s partially just because I’ve been seeing stuff and been able to trend where it’s going to.

And part of the thing that the ChatGPT did is it made the magic of current foundational models, large language models, available to people in a way that they hadn’t been available before.

And so, you get this like, “Oh my gosh, this is amazing.” And then, of course, that begins to show the lens to the future for a bunch of people. And then you get the classic, tidal wave of interest, both of where I generally am, which is this amazing place to where it can get. And then obviously, people who are always concerned and fearful about change, you get that too.

SAFIAN: So to kick things off Reid, we have Masters of Scale Executive Producer Jordan McLeod, who has prepared a special question for us written by ChatGPT.

McLEOD: My prompt for ChatGPT was, “Write a question about using AI to scale companies that a business journalist would ask an iconic entrepreneur who has founded a social network and an AI company.” And so here’s the question that it pumped out.

“Okay, Reid, in your experience as the founder of both successful social networks and an AI company, how have you leveraged artificial intelligence to drive efficiency and scalability in your business? And what advice would you give to other entrepreneurs looking to integrate AI into their operations?”

HOFFMAN: A couple of notes on the question, and then I will answer the question. One is: I think it shows some of the current limitations, although I think the GPT models will be evolving fast. And this is the current limitation, which is, it’s a fairly generic question. Bob in his sleep would come up with 5 more fun spins and other kinds of questions as ways of doing it. 

Two is that I do think that one of the lines of sight that I’ve seen from these foundational and large language models is that there will be a co-pilot for every profession. There will be a tool for every professional function between useful and essential. Last year, I said within 5 years. I was giving myself time. I think these tools will be within the next 3 to 5. Some of which already exist, like the Microsoft Copilot.

And I think that gives you some lens to how AI will also be used within various kinds of social networks, whether it’s suggesting or modifying, commentary, reply, or what you see in the newsfeed, which machine learning techniques that resemble current AI have been being used. Obviously, a lot has been used in advertisements. And people tend to beat too much of an alarmist drum about that because they go, “Oh, it’s manipulating you.” And it’s like, “Well, it’s trying to find the thing that’s most relevant to you.”

And actually, in fact, I’d rather see ads that are relevant to me. And it’s not manipulating me like subliminal advertising. It’s just simply saying, “Oh, you look like the kind of person who likes reading philosophy books. Here’s a philosophy book rather than a how to do cross stitch.” And that’s handy. And all of that is already present in social networks.

SAFIAN: Thanks Jordan. One of the things that’s interesting to me is that ChatGPT came out of a smaller, newer, less established business. The bigger players, like Microsoft and Google and whatnot, they’ve had AI products that they’ve been working on. Some that they release, but a lot that they’ve been working on more quietly. But it’s almost like no one wanted to take the risk of putting something out there with flaws. And now that ChatGPT has broken that seal, the gloves are off.

HOFFMAN: Well, I think it’s this classic set of things around: start-ups can take risks. Obviously we’re living in a very tech-lash moment. I’m endeavoring intensely, to get people to start thinking about how they can build such a great future and to stop trying to enshrine the past against the future and say like, “Oh, no. We can’t allow this ChatGPT thing to happen.” It’s like, “How do we use it?” But of course in the environment of tech-lash, what has happened is people say, “Well, you can get it to say something bad.”

And the answer is, “Yeah, you generally can. But by the way, you can also get Google or Bing to get to a bad search result too, or a YouTube video.” It happens. And boy, we’ve been talking for a long time about misinformation and issues in social networks and how to increase the level of information versus misinformation. So it’s: imagine the future where we can get to and where we’re going that’s so important on this stuff. That’s the reason that it’s important to take these risks. And the larger companies weren’t taking these risks. Although by the way, credit to Microsoft, which saw this stuff and started building out products, such as we began to see wthe beginning of the Bing announcement and Bing release earlier this week. And by wrapping in Bing, we’re getting factuality. We’re making it much more directed and helpful that search is normally bad at. Product reviews or travel or other kinds of things. We’re going to take some slings and arrows, but we’re going to get out there, get the product to market and iterate on it with the substance of trying to make it a great magical thing for most people, while you’re always fixing stuff that’s suboptimal.

SAFIAN: The ChatGPT frenzy didn’t come out of nowhere. You helped set up OpenAI in 2015, but back then it was a not-for-profit. When and why did the expectations change to turn it into a for-profit? Because it must have been something that you thought would help accelerate the pace of impact.

HOFFMAN: Well, it’s still fundamentally a nonprofit, and even the structure that’s set up now kind of rolls up to an economic return because they get the capital in order to do this. They’re getting capital, and capital wants return for that. And there’s a whole arrangement with investors, but it still fundamentally rolls up to: “How do you create AI that’s beneficial for humanity as a goal?” And setting that up as part of it. Now, as it’s matured, it now becomes a platform for a whole bunch of start-ups doing stuff.

It has in-depth relationships with Microsoft and other companies for doing stuff, all of which brings commercial drum beats into it. Each of these becomes its own hybrid unique beast, and OpenAI becomes its own hybrid unique beast and where it ends up. But I do know that the heart and soul of the OpenAI mission is entirely about: how do we make this maximum beneficial for humanity? And the commercial stuff behind it is to power that versus power an operating margin or a stock price.

SAFIAN: So to be able to generate the resources to keep that mission moving forward.

Bing vs Google 

You mentioned Microsoft integrating ChatGPT into Bing, potentially posing a threat to Google’s dominance in web search. Meanwhile, Google’s talking about its AI called Bard and rolling out its AI powered services. Does ChatGPT really pose a threat to Google? Is that the area that you think AI is going to have the biggest business impact?

HOFFMAN: Well, I think AI is going to have a huge business impact in search for a couple reasons: So, one is categories like product search get fuller and fuller of ads and junkier and junkier search results. And one of the things that you have an opportunity to do is to say, “Well, as opposed to giving you a whole bunch of links of which a bunch of those are paid links on advertising,” which by the way is sometimes a useful search result, but that’s not necessarily what the people who are paying for them are fully trying to achieve.

And as opposed to all lots of links, it gives you an actual answer. And that’s obviously one of the things that grew Wikipedia because what people are looking for is, “No, I’m not looking for 10 blue links. I’m looking for something that has a substance of an answer of what I’m looking for.” It isn’t some SEO engine to give me vanilla stuff. It’s like a personalized instantaneous Wikipedia page. It has some real accuracy.

Now, I think there’s a bunch of other stuff that will also be interesting. I think that AI will touch every single profession, not just the, “I happen to be using search to do research or figuring something out.” You already have Microsoft’s co-pilot for engineers. That’s the language that I’m using to say whether or not you’re a journalist or a podcast producer or an investor or a graphic designer or a lawyer or a doctor or a small business owner.

In each of these cases, there will be a one plus AI co-pilot tool to make you more effective as a professional. So that will be another whole swath. 

I was talking to a musician, and I said, “I’m going to tell you a little bit about how AI is going to be transforming your world. And the first 15 to 20 seconds are going to terrify you, and I hope by the second minute you’re going to be curious and intrigued and delighted.” And he said, “Okay.”

And I said, “All right. So right now, I have access to non-public programs that could go create lyrics and music and so forth that’s kind of in the style of John Lennon. And it won’t be great. It won’t be the, ‘Oh my God, Imagine,’ or other things, but it will be something like, ‘Oh yeah. Yeah, I could see that John Lennon could have made that.’” 

Then he said, “Okay, I’m terrified.” 

“Yeah, because you’re thinking, ‘Oh my God, I’m not needed anymore.’

But say for example, you were John Lennon, you had this tool. You’d say, ‘Well, I want to create this song about imagination and connectivity and mutual love and so forth. Oh, well, I really love the bits between the second 10 and second 20 and the minute mark and the minute 15 mark. I’m going to take those bits out and I’m going to make something much better.”

So then of course, he’s like, “Oh my God, I can create so much better now and so much faster and in different ways. So when do I get it?” Well, I don’t know, maybe next year, maybe the year after. I’m sure people will be building them. But that’s the kind of thing that we’re kind of getting into.

Then there’ll be the AI functioning behind the scenes, operating on tech, like the stuff we were talking about earlier in social networking, like relevancy for news feeds, or advertisements. These kinds of things will all be part of it. And that’s just the beginning. That’s just what we can see as the curtain begins to be drawn back and a little bit of light comes in through the window.

How entrepreneurs should be thinking about AI

SAFIAN: So Reid, how should entrepreneurs, business owners and leaders be reacting to the opportunities and risks that AI presents for them right now?

HOFFMAN: Entrepreneurs have to be skating to where the puck is going, not to where the puck is, because it’s kind of that build out of the future. And so they need to do the same thing here with AI. There’s obviously places where you go, “Well, the AI is not that central to this, how I’m doing it. And it’ll be changing for how professionals operate, but my thing won’t…” Fine, but if it is, the question will be: “Okay, how do I also skate to where the puck is going?” And in the competitive world that is business, that is start-ups, it’s, “How do I skate fast and accurately to that?”

SAFIAN: And so if I have a smoothie shop business, maybe it’s not my first concern is AI, but if I’m creating educational resources, or medical related things, it may be really something that I need to become more comfortable with.

HOFFMAN: Yes. And there’s a whole range. And even if you’re like the smoothie shop, you might be thinking, “Well, should I look at whether or not I have this as part of my answering machine? Is there a service there?” And obviously may not be so essential to building the smoothies. Or, “Maybe I should go on a ChatGPT and be creative about some interesting new recipes,” and that could be different. But it’s thinking about what are the things that I could do that I could use to skate where the puck is going? To be inventive and creative in the future?”

SAFIAN: Use the tool to differentiate yourself.

HOFFMAN: Yes.

SAFIAN: Where can you do that?

HOFFMAN: Yes. For example, I’ve never run a smoothie shop, but I’m already starting to think about, “Well, if I was running one, what if I had a special flavor of each day and how would I generate that?” Oh, I could use this to give me a bunch of ideas. Or maybe I want to experiment with a couple of marketing slogans on the local radio. Well, I could play with this and experiment with some marketing slogans. I mean, there’s a bunch of stuff to do.

SAFIAN: Yes. And I guess it becomes much more efficient if you want to do personalized outreach to your customers or your community. There are much more efficient ways to generate all of that through the communication tools of things like ChatGPT.

HOFFMAN: Yeah.

SAFIAN: When a new tech emerges, it often disrupts the existing tech giants. But I’m wondering if this is an area where the giants end up being the beneficiaries because of the resources and the data required to build and operate AI processes. That smaller and start-up businesses might be more vulnerable in the long run. Now obviously, OpenAI is a startup, so I appreciate that it started that way. But if you’re a smaller business, do you have to wait for one of these bigger businesses to create a tool like the ones you cite, like a co-pilot?

HOFFMAN: Well, with some knowledge of both the OpenAI and Microsoft plans, they are making APIs available to start-ups. They’ve done that already. It’s part of the platform business of what they’re doing. And there’s a number of my own start-ups that are already building on them. Tome, which is doing PowerPoint presentations. Coda with Shishir Mehrotra who we’ve had on Masters of Scale, is doing stuff around how this integrates into a mobile and internet native docs as powerful as apps kind of tool. And part of it is because there’s so much surface area to invent and amplify. The large companies will only be able to do a few of them that are most central.

So I think the answer is it will both benefit large scale tech companies, but it will also benefit a bunch of start-ups across the whole field. And I think there’s also a bunch of start-ups that are also doing their own fundamental work. Inflection is one of those, Adept is another. So I think there is fundamental work that is happening that will also create more space and opportunity that will also come from startups.

SAFIAN: Okay, we’re gonna take a break. We’ll be back right after.

Before the break in this special AI-themed episode of Need to Know, Reid and I talked about what AI and the ChatGPT phenomenon means for big tech, smaller businesses, and entrepreneurs.

Now we dig into the hype around AI — both for and against — some of the technical challenges and costs of AI tools, and what may be coming out of China.

Legitimate risks or overreaction?

You’re clearly super excited about this area. Not everybody is as excited. We’ve seen schools trying to block ChatGPT and creatives expressing concerns. There are a lot of threats that people ascribe to this technology. Do you feel like those risks are real? Are people overreacting?

HOFFMAN: Look, there is transformation, and with transformation comes stress, uncertainty, unsettlement, risk, pain, all of that sort of stuff. And I am deeply sympathetic to all of that. But it’s also when cars got introduced, it was like, “Well, okay, the whole horse and carriage business is going away.” When calculators became cheap and readily available, there was a huge, “Oh, this is going to cause people’s brains to erode because they won’t be able to do math anymore. They should still be using the slide rule.” And it’s like, “No, actually in fact, change is how we make progress.” Change is how we make much better and much bigger things.

Doesn’t mean it isn’t painful getting there. Doesn’t mean it doesn’t change people’s jobs. Doesn’t mean that people who are older and are less nimble and less interested in adopting new technology don’t feel at a disadvantage to the younger people who are going out and adopting this technology and making new things. And that creates uncertainty just like in education. It’s like, “Well, but wait, we have this college application process and we have this essay judging process and this means that we’d have to change.” And you’re like, “Yes, it does.” Full stop.

But as opposed to going, “Oh my god, change is terrible. It’s like, “No, how do we use this to get much better thinking and learning and critical thought and ways to amplify it?” Let’s say that at the speed of what we’re going, every cell phone can have an AI tutor and an AI doctor on it. If you delay that by five years, think of the under delivery of quality of life for every citizen that you are doing by delaying that. Is it your responsibility to delay that or to get that AI tutor and that AI doctor on every phone. And obviously, “Whoa, what if the doctor gives a bad diagnosis?” Well, by the way, doctors do give bad diagnoses. It does happen. What you want it to be is a good frontline. And by the way, if the person can’t afford a doctor, is uninsured or something else, having at least something that says, “Look, you should cross-check this and all the rest, but here’s something to look at,” could be super helpful.

And looking at that opportunity about what we can possibly do here is I think a moral responsibility. That’s what we should be really working towards. I’m not saying that there aren’t negative things that we should work against, but oh my gosh, should we get to the good things as fast as we possibly can.

SAFIAN: When we had legendary investor and Bridgewater Associates founder Ray Dalio on the show, you and he had a very spirited exchange about the upsides and possible downsides of AI. It didn’t make it into the episode, but the whole exchange is available to Masters of Scale Members. Let’s have a listen to a small section of it:

RAY DALIO: I’m a guy who has found that making algorithms has been so fabulous in my decision-making, and I really believe it’s totally great, right? But I have a belief, when you have artificial intelligence applied to things that might be different in the future and in the past and you get algorithms that you can’t really understand, they don’t give you good understanding, and you bet on them, you’re going to crash. And since we’re all connected, that scares me.

Reid Hoffman addresses Ray Dalio’s concerns about AI

SAFIAN: So Ray’s argument was that as AI proliferates, and it’s applied to more new things, you may get algorithms that reach conclusions in ways that we can’t track, can’t decode. And if we then rely on them, we add a layer of risk that, as Ray puts it, you’re gonna crash.

Ray isn’t alone in this sort of caution. Lots of folks are conflicted about AI. For those that take a similar position to Ray’s, what’s your response?

HOFFMAN: Look, I do think that it’s a risk factor when you have things running in systems you don’t really understand, which by the way, permeates our lives. The market is a very rich algorithmic system, and we don’t have good understanding, and that’s part of the reason why we end up in surprising phase changes, like the credit default swap scandal, and other kinds of things as ways of doing this. 

SAFIAN: Yeah, we don’t understand the economy and inflation either.

HOFFMAN: Yes, exactly. So we do live already today, biological systems, drugs, other kinds of things, and all kinds of complex systems that we kind of approximate, and then we kind of have fallback plans for when it breaks and what we do. And these will be some additional systems like that. Now, my hope is with these systems, we’ll also figure out how to use them to have our fallback systems. We’ll have to use them to understand these other complex systems and so forth. But I don’t think it’s because we’re only going to have simple systems. Maybe a more simple parallel is my 1972 Volkswagen Beetle. I used to be able to repair some with my Swiss Army knife when I was driving it around, not so much true anymore with the Honda that I’m driving or the Tesla that I’m driving.

You have to take those in. They connect it to the computer system, and it goes, “Oh, that’s the problem.” And that’s a complex system that does that, but you can design them and test them and do a whole bunch of stuff in order to make it work. So that’s the reason why suddenly was like, “Nah, Ray, you’re wrong.” Not to say that you’re not wrong that there’s an issue here to pay attention to and to manage, and it adds a risk factor, but to say that the goal is to have everything that’s kind of a simplicity form that most people can understand in its baseline fashion. That’s not where everything’s going, including the economy which he operates in.

The hype cycle around AI

SAFIAN: When new tech waves come like this, we tend to get super excited and then we sort of go over our skis and we get a little bit of hype in there. We certainly saw that with crypto over the last year. Do you worry about a hype cycle with AI?

HOFFMAN: Yeah. Look, I tend to think there are definitely times where the hype cycle gets out of control and promises things that aren’t going to happen at all in this tech window cycle, et cetera. We’ve seen that multiple times through virtual worlds and metaverses, I think I’ve lost track of a number of. “And now we’re all going to be in a virtual world,” kind of claims from the very earliest days of my own personal tech career. So there are times. On the other hand, there are also times where you say, “Look, the exact thing…” For example, I do think that there’s some places where there’s drum beating about AI advancing science, taking large levels of risk and so forth, that when you go and play with the current tools, you see both those positive options and other negative things aren’t currently in the cards.

I think there’s histrionics in both ways and that hype can be bad. On the other hand, the fundamental transformation of every professional job and every industry is full of professionals is I think line of sight visibility over N years. I think that’s one of the things that I don’t think we are yet overhyped on.

Lightning round question: Artificial intelligence fills you with hope or dread? Pick one.

SAFIAN: From the beginning of Masters of Scale, you’ve asked your guests about their feelings about AI in a lightning round series of questions that Masters of Scale Members get to hear. Does AI fill you with hope or dread? For a quick refresh Reid, let’s hear a few of the answers we’ve had to that question over the years:

HOFFMAN: Artificial intelligence fills you with hope or dread? Pick one.

HAMDI ULUKAYA: Hope.

SAEJU JEONG: Hope.

AJAZ AHMED: Hope, because it’s just a tool.

CINDY MI: I think it’s hope.

GWYNETH PALTROW: I’ve had too many conversations with Elon Musk about this. So, dread.

ARIANNA HUFFINGTON: It could be either, and it’s entirely up to us.

STEPHEN HAWTHORNTHWAITE: Oh. I’m going to have to go out on a limb here. I’m going to say dread.

ROTH MARTIN: I want to say hope. Hope.

ALEXA von TOBEL: I think we will make the decisions together as a humanity to make it be filled with hope.

NEIL BLUMENTHAL: Hope.

DAIVD GILBOA: Mostly hope and a tiny little bit of dread.

STÉPHANE BANCEL: I want to cheat. I love it, but I’m afraid of it.

LINDA YATES: Hope if in the right hands.

REED HASTINGS: Hope. It’s inevitable. It’s coming. Let’s embrace it and get good at it.

RAY DALIO: Dread.

ELLEN KULLMAN: Oh, definitely hope. Think about how technology has changed the world and my parents’ life, and their parents’ life, and my children’s life is going to continue to change. That’s going to be a positive part of it, I believe.

PHAEDRA ELLIS-LAMKINS: Ah, pick one. Ah, one. Hope. Hope. Hope. Hope.

SAFIAN: I love that. A range of answers there but, in general, I think it’s right to say our guests tend to ultimately fall more on the side of “hope.” Why do you think that is?

HOFFMAN: We get a higher balance of people saying hope on Masters of Scale because we have entrepreneurs and people who create the future and go do that. So they’re like, “No, no, we know how to shape this, and we’re going.” I think the broader populace and the commentators have a higher answer of dread because they don’t feel their hands on it as much. They don’t feel the creation of the future. They worry about change, et cetera.

So I think the thing to do is think about, well, what are the things we need to do to create more utopia and less dystopia? And it’s never only positive impact. There will always be some negative impact too, but the goal is to say, “Well, there’s 10x plus the positive impact or the negative impact.” Maybe 100x plus. That’s part of the reason why I talk about the AI tutor and the AI doctor because, well, those are good. He was like, “Yes.” 

The financial costs of AI

SAFIAN: What can we say about the resources required to run AI? Sam Altman at OpenAI tweeted that the computer costs are eye watering. I think that was the phrase he used. And the data behind it tends to need manual cleaning sometimes from humans. Will this cost of AI be passed along? Or do you think that over time that cost will, as with many industries, certainly in tech, will come way down?

HOFFMAN: Well, the short answer is both. I mean, when you develop something new and big and expensive, say for example, a personal computer, it starts out super expensive. I remember when it was like, “Oh, you have a PC? Can I come use your PC?” And obviously one of the things that’s great about competition and all the rest of this stuff for market share is everyone’s going to be trying to parse this stuff to get as much breadth of market participation as possible.

Let’s say for example, in whatever way it is, it costs you an extra buck a month in the search category or two bucks a month in a document creation category. Well, what’s the amount of time that it would need to save you to make that worthwhile? If it saved you 3 hours on your searches per month, that worth a buck? The short answer is to many people, the answer is yes. Absolutely. And we say, “Well, but that’s a first world answer.” It’s like, “Yeah.” And as the prices go down, that will become everything just like everything else.

Part of what the internet brings and mobile brings and the cloud transformation brings is that you can get the product distributed to a lot of people much more quickly, which then of course means when you have a much broader customer base, your pricing doesn’t necessarily start so astronomically.

Reid Hoffman on Baidu’s Ernie Bot

SAFIAN: The other thing I wanted to ask you about was Baidu’s Ernie Bot out of China. Are they legitimately in the AI arms race, if not for English language search, then for AI development? Do we even know?

HOFFMAN: So Baidu has all of the talent to do this kind of thing, and a lot of stuff has been published as papers so they can look at the papers and they can do their own work and they make it. So I would presume that it is a real thing.

And I haven’t yet gone and called my friends who are native Chinese and deep technologists and said, “So, you’ve seen ChatGPT, did you see this? Anything that you could pass along as a learning or interest or comparison with Ernie Bot?” But I’m certain there will be. You know, personally, China is one of the places where I most learn new parts of the theory of the game from technology, whether it’s go-to market or product or business model or other things. And so along with a number of Silicon Valley people, I pay a lot of attention to the inventiveness and the creativeness and the hustle that you see in Chinese technologists and entrepreneurs.

SAFIAN: Well Reid, we’ve covered a lot of ground on this topic, but at the same time, I feel we’ve barely scratched the surface. Fascinating as ever, and thank you so much for your insights and your sharing.

HOFFMAN: Always fun, thanks Bob.