The Exchanges

Q So, um, let's start here. There's there have been headlines that, um, of the announced AI data centers that are supposed to come up, something like 50% of them are actually being built. Is that the case? And if so, why? Anissa, do you want to lead us off?

A Sure. I totally believe that statistic, and I think it might actually be higher if you include announcements because of all the planned data centers that are underway. I think it's highly likely that many will be delayed due to higher costs, how hard it is to get labor. But then the number that I'm keeping close track of is announced projects versus actually projects that are being built. And I think we've seen some You know, pretty crazy announcements from companies like OpenAI with all of these different 10 gigawatts, six gigawatt projects. And those are numbers that I'm paying a lot of attention to because I think we need to sort of back into them and say, OK, if you wanted 10 gigawatts by this date, how many do you have today? Was that a real commitment? How firm is that commitment? But on on projects that are actually getting built, I do think, you know, 50% not really getting done on time is is, you know, what I would expect.

AI assessment note: “I totally believe that statistic... delayed due to higher costs, how hard it is to get labor”

Q whether, whether the margin of the business can be maintained at the current prices. And I thought, OK, well, forget about it because these labs have such an economically valuable tool for us that they'll raise prices. But we might be in the moment where they're going to get into a price war because OpenAI is rumored to be potentially dropping prices. And so what do you think about that?

A I think we're absolutely right on the precipice of that. I think you actually saw in the audience here by comparison, we saw every hand go up when you said would you be willing to pay double when we were together three months ago in April. And when, when Alex asks if people were willing to pay four and five times as much, there was still a quarter of the hands in the room that were up. And we're talking a room of about 200 people roughly, right? So it was a lot of people was a good, you know, good, good tea sample, so to say. I contrast that to what we just saw right now, and I think it's a fairly, you know, even distribution, similar subset of people. And the reality is there's there's more skepticism of value that they're getting from it, especially when you start layering on the access and the capabilities associated with Some of the models that are still per seat, as well as some of the open models, which you can get access to and, you know, for free, you can do a lot of really cool things.

AI assessment note: “I think we're absolutely right on the precipice of that.”

Q Definitely. Well, we, there was a guy named Ben on X who said, I will mail Anthropik an original copy of my long form birth certificate if they will enable Fable for me again. I sound like those lunatics who were obsessed with Foro now. Will you take Ben's long form?

A I don't know that we'll take his long form, but it has been interesting. I mean, Fable was only available for a few days, but I've definitely, every time I've tweeted since then, they have not read whatever I was tweeting and they've mostly been like, Bring back Fable, which like an Instagram where you got rid of Gotham. Do you ever got them the filter? This is like, and then for the rest of like the next eight years, all I heard was bring back Gotham. So it struck an earth, but Fable will, uh, will come back before Gotham did. But yeah, it's, it's clearly the folks that have gotten to use it and started incorporating it. It's actually really interesting. Um, I've learned to not really trust day of or even week of model reactions. You don't really know until you've put it through its paces. And so like, I almost just, Completely block out the noise in the first couple of days of any new model release, because I don't know, everybody has maybe their like toy, like example thing that they like to do with a new model, but it's hard to actually put it through its paces until you've actually had real work done with it. And I think people were just starting to do that. And then, you know, we had to sort of pull back Fable. But I remember in December when we put out Opus four, six, it was like this interesting time where everybody went home for the holidays and a lot of people had tha…

AI assessment note: “I don't know that we'll take his long form, but it has been interesting.”

Q to see this continue, agents can't be limited. They have to be able to operate autonomously and spend all those tokens and be effective. So what are the limits that you're seeing with agents today? And do you think that the like, if we could extrapolate a little bit, it means that we're going to see some more speed bumps as the labs try to roll this technology out further?

A Yeah, I mean, you use the term limit. I think it's, um, I think it's, uh, a function of both risk tolerance, um, as well as cost tolerances. And then finally, like, what is it you have an expectations of these things doing on their own? And so the limitations are actually in all in all three of those areas that are coming through. Um, you know, from a risk tolerance point of view, I think people are saying, wait a second, I'm worried that the agent without some level of, you know, Call it control or governance around it. It could go just about anywhere. And what does that mean within my organization, depending on what access I give to it from a data perspective? Um, you know, I would say from client data or whether it's, you know, it's code itself and what can it do to change code? If you ask to do one thing in one area, will it simply think that it needs to do that everywhere else? And there's a, you know, we'll say the ability to extrapolate on a single point and like what control exists there. So that's the first limit. The second limit is on, like I said, on the cost experiments, variance. We went through there a second ago, which is, hey, look, like there are just things you're not going to want it to do because back to the MIT study, there might be things that humans can do not only better, but more cheaply, especially now, depending on, you know, if you're, if you have a…

AI assessment note: “the limitations are actually in all in all three of those areas”

Q it wasn't for this guy. And it's a picture of Boris Terny, the person who created Claude Code. And so for companies that are going to build on top of Anthropik technology. You know, they're going to wonder, do I want to partner with Anthropik or is Anthropik going to go ahead and and build the product that I'm going to want to build potentially even after partnering with them?

A Yeah. I mean, we'll take the like agenda coding side and I think the broader sort of aspect of, of, you know, being both a platform and a product I think is really interesting. When we take on projects, it's the goal is often to sort of Push that area of the industry forward. So, um, you know, there were AI coding editors and some of them were really good and, you know, uh, but nobody was quite thinking about it in as sort of freeform a way as we got to think about it with cloud code. And now a lot more products have that flavor than I think would have otherwise. And so I think if wherever you can call me out on this, Alex, like if we're ever entering an industry where like all you're doing is the same thing everybody else is doing, but like you've got the anthropic brand, I feel like that's a bad use of our time and a bad use of our Either labs or product team time. Like if we're going in somewhere, it should hopefully be to say, all right, we think that the direction of travel is this way. We can build a product of that. And then, by the way, there's no world nor should there be a world where like all the products are in topic products. That would be a bad world. Right. So like that is hopefully either creating new space for companies or sort of showing the way where other products can incorporate.

AI assessment note: “hopefully either creating new space for companies or sort of showing the way”

Q Here's an interesting thing that Larry Ellison said recently. He talked about how AI models are rapidly commoditizing because most are trained on the same public internet data, which I think we agree with. He said, This makes exec exclusive proprietary proprietary data sets a real competitive Moog. What do you think about that?

A A hundred percent. I think, um, what is truly novel and unique Is an enterprise's data. It's their data. It is a living record of their enterprise, its decisions, its actions, its operations, and through it, you can get, you can realize enterprise context for that bespoke organization. This is not information that's available to the frontier labs. This is not information that's, that's available beyond the boundaries of your organization. Now, if you can find the right mission critical data with the requisite quality and apply the models and the compute for the appropriate business case, that's when you can get transformative outcomes for your organization. And the good news is you have the data. But you need to ensure you have the appropriate technologies for governed compliant use of your enterprise content, your unstructured data, as well as your structured data.

AI assessment note: “A hundred percent. I think, um, what is truly novel and unique Is an enterprise's data.”

Q So Mike, what happens to the people then?

A Well, the people are all focusing on more important activities, right? I mean, the people, think about, think about some of the scenarios that we're talking about. We're talking about searching for information. We're talking about reading documents and checking to make sure that they're, that they're complete. That's relatively mundane activities. And The ultimate goal is to free up people to do more high value work in healthcare that spend more time with patients. I think we would all acknowledge that when we go to the doctor, we actually don't spend a lot of time with the doctor. Why? Because they're spending a lot of time doing paperwork. Um, when you think about, you know, any of the, any of the industries that we serve, um, be it, be it, uh, healthcare bank government, um, We, we, we do, we do a ton around government funded benefits eligibility. The, the, you've lost your job, unfortunately. Now you're applying for unemployment insurance. Somebody is assigned your case. Somebody is checking all of your information. Somebody is determining maybe you're eligible for other benefits that you hadn't considered. That's all work that can be automated. So that what? So that those people, those resources, that money can be spent on other services That could be much more beneficial.

AI assessment note: “the people are all focusing on more important activities, right?”

Q know They are not software margins. Traditional software margins. Like, the marginal cost of distributing a product is not effectively zero. So, like, we know it's gonna look different, but them chasing and driving this in the wrong direction, I think, could be really bad. Do they, do you think they're gonna do it, Juan? And do you think it would be good or bad for them if they do?

A It'll be fun to ask Greg Brockman about it next week at the summit. Um, so I will do that. Um, I think they will. I mean, I think that there's, if you think about it strategically, right? Anthropic has been winning. It's also, it's like part of this is about lifetime value, right? Anthropic has been winning over a lot of these enterprise customers with its offering. And once you win these people over and you get them set on, on, you know, your system, they're not going to want to go anywhere else for a while because they're going to build that habit, right? And because if it's working for them, then why would they go somewhere else? And, um, and so I think that OpenAI probably sees it as, well, if we drop our prices, we're going to lean on our strength, right? And what's their strength? They have all this infrastructure. They've invested in more infrastructure than Anthropic. Dario called it Yoloing, right? What they're doing on the infrastructure side. If you've built the infrastructure, why not lean on that as a strength and try to win these companies over, um, and get that lifetime value. Uh, that will be difficult for Anthropic to get over time.

AI assessment note: “I think they will. I mean, I think that there's, if you think about it strategically”

Q interesting to hear you say, okay, use case number one is the speaker. Use case number two, photo video. Um, is the reason why AI is below that because, is it because the AI isn't good enough yet? What do you think is holding AI back? I mean, if you think about the power that people say AI has, you would imagine that it should be use case number one.

A The potential of AI is undoubtedly way higher than anything else right now. Just, if you just look at the sheer rate of improvement in the industry, it is abs, it's mind boggling. I think for a lot of people when, for a lot of people still, when they're looking at AI, it looks like a blank canvas on a wall. It's like, it's like if I handed you a piece of paper and a marker and said, all right, make me something great. Kind of don't know what to start with. Um, and so part of what we're doing while we start with auto capture and we'll have some other, um, things launched shortly that are very relatable. It's just trying to give people easy to understand, easy to use things with AI. And over time, you'll be able to do more and more complex, uh, and advanced tasks that really take advantage of the underlying capabilities. I think people are still learning how to best use AI. I mean, it started as in general, it started as a tool to help you write things. And I think now you're seeing people are coding with it in new and crazy ways. You're starting to see agents and agentic behavior take off where you can give the AI a goal and it will go accomplish something for you while you're not actively prompting it. So I, I think we're seeing that progress happen quickly. And we believe that wearables will be the best form factor, uh, eventually for the AI. Because they're always with you, t…

AI assessment note: “when they're looking at AI, it looks like a blank canvas on a wall”

Q So, is your thought then, again, if, just, I'm trying to flesh out the product vision here. If you have a pair of glasses like this with you while you're, while you're working, right? Which I guess is the case for you right now. You could just say something to it, and then some agentic Engine on the back end will be able to do these use cases for you.

A That's, yeah, I mean, that's a pretty good articulation of it. Um, I mean, I think of it in two ways. One is, and it's in two directions. One is, um, as I'm going through my day, and there's something I, you know, a lot of, it's funny how many people in tech decide that the thing that they need to invent at their next startup is a, like, a task app to just help you, you know, track the things that you need to do. And I think instead, as things come up, like, oh shoot, I forgot to, you know, I forgot to send out the email with the details for the game, or I forgot to send out a birthday invite for, you know, my kid's birthday coming up, you can just say, hey, I forgot to do that, draft it for me, and I'll look at it tonight. At some point, you can say, send it on my behalf, but I think we're still kind of in a draft version. I think that's, that's one end of it. The second is, I like to describe, it's, You know, imagine if you had the perfect person sitting right next to you who will whisper in your ear, At the right time. So, like, you know, my wife is very good at, you know, elbowing me. Like, if I'm about to say something that I'm gonna regret, my wife is really good at letting me know that I'm about to do that.

AI assessment note: “That's, yeah, I mean, that's a pretty good articulation of it.”

Q and codex apps, Where, which has driven so much of the growth in AI right now, which is sort of, uh, applicable for coders today, but over time, if the AI labs see it, you know, see their way, prove out, everyone will use these type of apps. So I'm curious to hear your perspective on this. Um, is this as big of a liability for Google as I'm imagining?

A I mean, I do think it's a problem. You even heard Sundar Pichai was on, um, you know, with, with Casey Newton and Kevin Ruse on their podcast, and he explicitly said, like, just straight up, like, um, that they're a little bit behind in, in coding, you know, with regard to AI and everyone knows that this is sort of the forefront where you need to be, not only from a actual coding perspective, but because a lot of people of course think that this is what is the key for sort of the agentic use cases going forward. Um, And, you know, potentially self, um, you know, these models that can teach themselves and whatnot, and we'll see how that plays out. But still, it's clear that everyone recognizes from open AI, which realized they had to sort of, if you don't want to call it a pivot, they had to reorient the entire business right around. Yeah. Building towards the super app and bringing in a codex and, and putting it all together, which they still have not done yet, but presumably were We're closing in on that happening. Um, and now, yeah, Google recognizing and acknowledging that they're not where they need to be with regard to coding, and there have been reports, you know, that, that everyone from Sergey Brin on down is sort of, ah, all hands on deck, literally, to make, make sure that they can sort of catch up in this world, because of course, they're Google. They should be. They…

AI assessment note: “I mean, I do think it's a problem. You even heard Sundar Pichai”

Q more than five years away. This week, the week that we were recording, um, he said, when we look back in this time, I think we will realize that we were standing in the foothills of the singularity. What do you think that statement means? Um, and, and what do you think about the fact that we've gone from five years till AGI to foothills of singularity in a year?

A I don't know exactly what that metaphor means, but I think he's indicating it's coming faster than he thought. Um, of course it's jagged. So it's not like it'll get smarter than people or as smart as people at all things at exactly the same time. It's already way better than us, the general knowledge. These AIs know thousands of times more than any one person. Um, it's way better than us at playing games. It's already way better than almost all of us at math. Um, and it may soon be better than all of us at math. Um, it's still worse than us at some things. So it's, it's very jagged. Um, so the whole concept of AGI that it's going to be equal to people at everything all at the same time doesn't really make sense to me. It's going to be better at some things, worse at other things. But right now, I would say we're at about, we're close to AGI, because if I ask a chatbot, I can ask it any question, and most of the time it'll answer at the level of a not very good expert. It'll be much better than me at anything I don't know a lot about. So in that sense, we've really reached AGI.

AI assessment note: “I don't know exactly what that metaphor means, but I think he's indicating it's”

Q In, in your estimation, um, you talked about how it's moved faster than you expected. Um, Uh, what do you think has enabled it to do it? Is it techniques? Is it the fact that there's been this data center rush? And what didn't you anticipate about the progress here?

A Um, it's a combination. Obviously, there's been huge resources put into it. For most of the history of your neural network since the 19 fifties, there were just a few people working on them with modest resources. Um, over the last few years, we've seen, um, Hundreds of billions of dollars, maybe trillions of dollars put into AI. Um, so that's certainly one factor. We've also seen a lot of progress in the engineering. So without sort of major conceptual breakthroughs, the engineering has become much more efficient. So things that were sort of inconceivable a few years ago, they can now run. Um, We've also seen new ideas, but, but mainly since Transformers, it's been much better hardware, many more resources, um, better engineering, and many more talented people. So, 20 years ago, there were a few hundred, few hundred people doing research on neural networks in the whole world. Um, now it's more, more like a million, I guess, I mean. There's lots and lots of people.

AI assessment note: “it's been much better hardware, many more resources, um, better engineering”

Q mean, who cares about your nine hundred twenty million active users or nine oh five or whatever it is like, and again, is the entire game going to be codex? Like, I don't, do you think, do you think they're going to still go, do you think they go into IPO with a dominant story still being Consumer plus enterprise. Even after they told us it's going to be enterprise.

A I think it's going to be enterprise. And I think again, and this is something I've brought up in all of my interviews recently is, you know, a lot of the growth of generative AI has been on the back of like novel uses that's gotten people interested, uh, in the technology and sort of established a baseline behavior, but hasn't exactly become the norm. Like, I think we're still searching for like, what is the key use of generative AI? And remember where we've seen these spikes, image generation, voice generation, and now this agentic stuff. The question is, what is going to be the mainstream use that everybody goes for and is intuitive and useful for everyone? And, you know, that we don't know whether the codecs and cloud codes are going to be that type of mainstream use or whether there are another flash in the pan.

AI assessment note: “I think it's going to be enterprise.”

Q Silicon Valley, um, kind of learned from you. The Obama campaign in 2008 told people hope and change. One, um, tech started focusing its messaging on how it was gonna change your life and give, give you hope. Um, and a lot of people don't feel that hope in their lives. Is it that that messaging is just falling flat on people and they're just saying, I don't believe you?

A Well, we obviously have a historic distrust of institutions and people in power today. So voters, and when I say voters, it's not just voters, citizens are disinclined to believe messages like that, that they believe that somehow I'm gonna, I'm gonna be, I'm gonna get the shaft. While you might profit. So I do think, you know, particularly around health and some of the breakthroughs that we're seeing happen with AI. And of course, you know, a use case that probably 60% of the country is engaged in now is to use these LLMs for their own health diagnosis and research and people getting great value out of that. So that's a place where I think you can definitely sell hope. I think it's going to be really important for the industry to lift up places, small businesses, medium sized businesses who are utilizing AI To make more money, to be more productive, to grow their business, but they're also growing their footprint. Because all these major layoffs that have happened at the major companies are seen by Americans. Like they're not necessarily reading the entire Wall Street Journal story, but they see it and it scares them. So I think there, there has to be more positive storytelling. I think some of the storytelling to date around how people are using it for fitness and cooking, that's okay. But to me, it doesn't address the core concerns people have. You've got to tell them, A, tha…

AI assessment note: “citizens are disinclined to believe messages like that, that they believe that somehow I'm gonna”

Q now. Uh, and the new numbers are coming in, uh, April, May employment, and they're looking pretty good. Um, which would suggest that maybe college grads are going to be okay. So where do you think we are on AI and jobs? And does this extend beyond, you know, Sam and Elon talking and Dario? Let's not forget him talking about how there's going to be like white collar bloodbaths.

A Well, I think, right. My, my sense is most of the announcements to date have been a blend of AI and a convenient way to downsize. Um, and I think that's across sector, across industry, and you see a lot of copycat announcements in that regard. I think when you talk to voters, most of them don't know somebody yet who's lost their job directly because of AI, so it's more of a concern, right? Now, all of us, I assume in this room, ah, you know, even those of us that don't work in an AI company are using it all day long every day, and we see the power of it. And things that we would have asked somebody six months ago to do, who's an entry level person, you don't do that anymore.

AI assessment note: “most of the announcements to date have been a blend of AI and a convenient way to downsize”

Q flights. Uh, I talked about it, uh, with marketing presentations, uh, and, you know, the week that we're talking, you have a, a new use case out where, um, Claude Cowork can be used for small businesses, including, uh, taking over QuickBooks and doing some bookkeeping. Um, where, where does this go? I mean, what do you think the broad roadmap, where does, where does the broad roadmap take you?

A We're thinking about a few things for quad code and for co-work. There's a few, few big themes. One is improving intelligence, and, you know, I, I think almost all of this is just the model. As the model improves, we can do more and more ambitious work. For coding, it used to be writing a line of code at a time. Now it's building entire features or entire products. For co-work, it used to be, you know, like, you know, it started pretty recently, but it was like, you know, making a document, and now it's things like booking flights, combining many tools, doing, doing your QuickBooks. Um, so this, this frontier is improving and moving just very, very quickly. We're also thinking about how to do longer running tasks. For Cloud Code, we recently shipped this thing called Auto Mode, and Auto Mode is, ah, essentially a replacement for permission prompts. Before, what we used to do is, whenever the model uses a tool, Cloud would ask you, is it ok if I use this tool? And, you know, usually you just say yes, and you get kind of tired of saying yes, kind of over and over.

AI assessment note: “We're thinking about a few things for quad code and for co-work.”

Q a world model, um, a, uh, LLM just doesn't have an understanding of the way that the world works and consequences and stuff. You use co-work to book how many flights, eight flights in hotels? Like, you must think that it has some understanding of consequences, otherwise you wouldn't have given it. Your credit card, which I presume you did. So what do you think about that argument in particular?

A I think from what I've read from folks working on, on a research at Anthropic, it is surprising the degree to which these models are intelligent. Because like you said at the beginning, the, the thing that they fundamentally do is they predict the next token. And so you think like, this is kind of like a stupid thing. Like how can this possibly lead to intelligence? But you know, we, we've actually published a lot of work about how the models are able to plan They're able to actually reason. Um, there was all these, like, very surprising behaviors that you actually wouldn't expect from a model that just predicts the next token. So, I don't know. I, I wouldn't discount it.

AI assessment note: “we've actually published a lot of work about how the models are able to plan”

Q the compute to people using IP he doesn't own. Which is the other side of the, you know, the praise he's been getting, right? The praise he had been getting by being the company that had locked down OpenAI and invested in them and, and given them, um, and had rights to their IP. So what did you think of the fact that Satya is framing this this way internally?

A Is he wrong? Do you think he's wrong? Cause I actually think my favorite part of this is I feel there's some tech leaders. You get a kind of, we all know what they're thinking in general, and then there's those who are just so savvy that they keep it close to the chest. And this is one of the first times I think I'm getting a genuine look as to how Satya Nadella really thinks. And again, he was prescient on this. I mean, he recognized exactly Where they were, where the shortcoming was that, and I love that better to be an investor and not even take all this execution risk. So he recognized this idea of like the more partnership element as opposed to just the pure investor element was a major risk to the direction Microsoft was going. And I think we're going to get more into where Microsoft is today in just a bit, but he saw it coming. Satya Satya is king.

AI assessment note: “he was prescient on this. I mean, he recognized exactly Where they were”

Q of record, so to speak, um, that handles all this stuff. And I think they, they might even prefer a world where, you know, that would just be the single app to rule them all. Um, you're orchestrating their models. So long term, aren't you sort of at least dependent on their benevolence to allow you to use these models, um, even as you compete with their core products now?

A Yeah, I, I think ultimately all these companies are platform businesses in addition to product businesses, and they, You know, they aggressively petition us to use their models. They want, they, they give us early access. They want us to run evals. Um, and so we, we have, you know, the exact opposite dynamic right now where they're, they're, you know, more than happy to take revenue from us. Um, and you know, they're the beneficiary of, of, you know, more consumption of computer credits as well. Uh, and, and I think they, You know, because they are all competing with each other on their platform businesses as well, and, you know, there's open source, which, which is, you know, you know, Continuing to push at the frontier, not necessarily at the frontier, but pushing at it. All those competitive dynamics are very healthy for us. Now, I agree with you. If we lived in a world where there was just one frontier model that was twice as good as the next best model, that wouldn't, that would be a bad scenario for perplexity. I wouldn't deny that. But, you know, Since this industry has kicked off, there's never been a moment where the delta between the, you know, the best model and the second best model was, like, more than maybe, like, a 10, 15% gap. And again, like, best model is, is probably, I shouldn't even be using that phrase because it's best model at what, right? There's differ…

AI assessment note: “All those competitive dynamics are very healthy for us.”

Q unlimited. Like you wouldn't do an unlimited electricity plan or an unlimited fuel plan. But for some reason, a lot of these companies have been doing this. Um, do you think that this is like a legitimate issue that she's pointing out that basically like we don't really know what AI demand is because it's been subsidized so heavily for so long? Uh, and if so, what's the answer here?

A So we, uh, at Perplexity, we've never subsidized paying users. So if you're on a, you know, pro or max plan, uh, you know, we're, thank you, you're, you're contributing to our success. You're welcome. Uh, the, uh, and, and, you know, we see great retention, so clearly folks are finding value there. Uh, and that's actually why computer credits are so important, right? So that as you have, Uh, because you can have a certain computer task, uh, cost you 50 dollars for, you know, say it's like video generation, and it's like long horizon running. You know, you can, one task can cost up that much, and then you have certain tasks that cost, uh, you know, five cents. Uh, and so there's no way to encapsulate all of that in a, you know, subscription product, right? So I, I think the mental model I would have is, AI is going to become a lot like Costco, uh, where you pay for the membership, right? And that gets you in the store. Uh, and that's actually the part of Costco's business that is, you know, the highest margin. And then you have, you know, everything you're buying in the Costco, you know, you have confidence that there's like a max margin, right? And those are kind of like computer credits, right? And it's, you know, some people go to Costco and they just buy the hot dog. Uh, and then, you know, you know, there's people who go and spend, you know, thousands of dollars every trip,…

AI assessment note: “AI is going to become a lot like Costco, uh, where you pay for the membership”

Q Okay, speaking of another area that may be disrupted, ah, consulting is a field that people talk about often as things like implementation and doing business analysis. You can prompt that in one shot at probably. You mentioned that your daughter is going to work in consulting. Did you bless that?

A Yeah, of course. Yeah, because there's going to be a period of time where somebody's got to help all these companies figure out how to implement AI, um, And not only implement it, but how to reformulate their businesses. Because how you do things in an AI universe where you're trying to be more productive, more profitable, more competitive, it's going to be completely different. The hardest challenge for those CEOs is, am I willing to blow up my business, um, knowing that my stock price could collapse in the meantime, and then I have to build it up. And working for a consulting firm, You know, you hopefully have a lot of proprietary information that will allow you to guide that CEO as they go through that process.

AI assessment note: “Yeah, of course. Yeah, because there's going to be a period of time”

Q They are. Is it over hiring and a bloated company that just needs to actually trim itself, which is I think very true for many, many of these big companies, especially over the last five years, or is it really like we need to cut costs so we can invest more in artificial intelligence?

A I think it's the latter. I mean, they've spent so much money on AI. They're spending on the data centers. The market is actually much less forgiving if you don't have margins, right? And they, they don't have, I mean, they have some ROI on the AI because it's helping optimize their creative stack. And I think there was a headline recently that they're going to pass Google as the largest advertising business. So they see that the results there, but they're not a platform that's sort of benefiting from this surge in demand for AI compute. They haven't built super AI super intelligence. Um, so they are, they are in a position where they cannot remain this bloated, especially as they don't have, um, the leading model.

AI assessment note: “I think it's the latter. I mean, they've spent so much money on AI.”

Q And it's like, yep, the whole job is trying to figure out how the generative AI systems work. And it's like, you're putting this out there, people are relying on them. And then, oh, I mean, I guess, I don't know. And along the way, you're, you're trying to like figure out how they work. You're trying to interpret them. Like, isn't that backwards?

A Yeah, it is, and something really interesting, it's, it's, it's a bit of a metaphor, so I'm not saying it's exactly the same thing, but we can really learn a lot from ancient Greece and ancient Rome, because our current, you know, we started this conversation by just pointing out how much we're relying on prediction, and we've always relied on prediction, but I think there are times in history when that goes up and goes down. I think this is a peak, and another peak was ancient Greece and ancient Rome, and if you were to interview an ancient Greek person and say, what do you think about the Oracle of Delphi? They would say, oh, it's, it's cutting edge technology. You know, it's, it's the best we have to make decisions. And how does it work? Well, we're trying to interpret it, right? And the same thing with astrology. It was a very technical thing about how, how to read the stars, how to measure the distance between the stars. So in, in a way we've seen this before, even though the technology is different, the political role is actually quite similar.

AI assessment note: “Yeah, it is, and something really interesting, it's, it's, it's a bit of a metaphor”

Q Different use cases. One is an air war, one is a ground war. What are the main things that you've learned watching this in action, and what do you, how do you think it changes, again, the way fighting might happen?

A Yeah, two different, you're right to point out two different scenarios. So, in Russia, Ukraine, you have a battle over territory, and so that battle over territory, where the lines are drawn, means that with the drone warfare, the robots are the front line, and the humans are back, and the idea is Well, why risk a human going in front if you could send a machine first and see what, you know, see if you could fight it that way. Still a lot of destruction, death there that's obviously sad and unnecessary, but I don't know how much more there'd be if you had a civil war style thing where you have, you know, humans on humans. In Iran, um, the drones, I think the lesson from that is that, um, the imbalance of costs Right? You have a cheap drone going against very expensive targets.

AI assessment note: “with the drone warfare, the robots are the front line, and the humans are back”

Q say about this, and I'd be curious to get your thoughts on it, is that you can look at the history of companies that have been deemed a supply chain risk to the Pentagon, It's very rare if unprecedented for a company like Anthropic to be banned that way. So why do you think it rise to the level, and do you think it merits this, like, fairly unprecedented action?

A Well, I mean, on one hand, you can't say that they have this mo, this cyber nuclear bomb, and yet we shouldn't be worried about how those capabilities enter and remain In our supply chain. Those two things are inconsistent, right? And I'm not blaming you. I'm just saying that if you believe they're going to cause 40% unemployment, if you believe that, um, these things have a capability that you put 50,000 geniuses in a data center, they're going to coerce the world. They could create bio and chem weapons. Of course the Department of War is going to want to understand and constrain, uh, you know, those things so that they don't do something unintended on our side, right? So these companies are talking about their things in apocalyptic terms, which make it necessary for us to judge the management teams, judge their actions, Look at the terms of service. Understand how they fit in our supply chain. This technology is like nothing we've ever seen, so you can't compare it to, you know, a chip from a foreign chip manufacturer that gets put in the supply chain. This is a whole different thing because of just what you said, is the power of what they're saying it's going to do, the disruption might cause an American life. Um, and we don't, if someone developed a nuclear bomb in their In their garage, you don't think we'd have anything to say about it. Yeah, well, of course we would, rig…

AI assessment note: “These companies are talking about their things in apocalyptic terms, which make it necessary”

Q So just talk to me a little bit about what you think this looks like in an ideal world. Like, this all works out. How does that change people's lives?

A It changes people's lives by taking away the busy work and not just doing that in a way where you're getting a ton more content that is mid, that's average, but you're getting a ton more content that is highly optimized and specialized for the way in which you work, in the way in which your company has set its mission and values, the way in which your company runs, and what your company level goals are. So that should be elevating the, the ability for the company to hit its outcome metrics. It's true, like, Key results versus, oh, we shipped a lot of code, but we actually didn't manage to sell the product, or we shipped like five new campaigns, but they were undifferentiated. So getting to the level of differentiation, getting the key results, elevating every human team member to becoming a tastemaker, those are the outcomes that I'm driving for.

AI assessment note: “It changes people's lives by taking away the busy work”

Q this progression where it was basically like you start with OpenAI because that's the most common one. Then you make your products interoperable so you can use any model within them. And then you graduate to open source so you can customize it more. So there have been advances on the customization front from the frontier labs. That have allowed you to build on top of them versus open source?

A That's a good question. What's happened to that? Um, so again, like, uh, our sort of maximalist thinking is that the frontier labs are going to keep innovating in the level of reasoning and capabilities of their models, and so trying to, uh, create these customizations or adding our own token weights Is not a good idea and is a waste of R&D resources at this particular point in time because of the rate of innovation that we're seeing. Like, why not just trust that they will keep innovating in, in their space with the funding they have and the quality of research talent that they have, uh, and then just basically evaluate which of the ones work best for our use case. Um, my perspective might change over time, but like in this day and age, I'm seeing like enough velocity coming out of them That it doesn't make sense to try and create, like, a separate path where I might just fall behind.

AI assessment note: “trying to, uh, create these customizations or adding our own token weights Is not a good idea”

Q So I've heard you say this a couple times that the CEOs may be downplaying the impact. Uh, I know they speak with you privately. Are they telling you things like saying, hey, Senator Warner, don't say this to other people, but here's what we think, or what, what brings you to that assessment?

A Well, what brings me is You know, the CEOs who are saying this in the AI space. And what I'm hearing privately from big brand name firms who are saying they're, they're cutting off or cutting in half the number of interns or first year hires. I even heard from a nationally known law firm that has decided to hire no first year associates. They're going to take a pause. And see how this works out before they even hire all these kids after they've done everything to get through law school and they got a job offer they thought with a big brand firm. And then it's just going away. Nothing they did. And because of AI and yeah, because AI and I hear, and I hear like so many companies that are midsize who say, you know, I had one guy the other day saying, you know, I had 23 people do this back office function. Now I got three. Isn't that amazing? And the thing is, we are not even collecting data on this yet. That's why I've got a, a bill with Josh Otley, very bipartisan, that says to BLS, Bureau of Labor Statistics, we need to start measuring this, um, and, and not just in terms of firms like a Jack Dorsey saying he's cutting 40% of his staff on, because of AI, and whether that's true or not, you know, we'll, we won't know for sure, but, you know, that kind, but also try to measure You know, jobs that would traditionally have been created, because my view is that this is going to parti…

AI assessment note: “what I'm hearing privately from big brand name firms who are saying they're”

Q years old to use our platform. Do you certify that you are? Some are even saying, like, I think OpenAI has been saying recently that we have technology that can determine You know, smartly, whether you're a minor or not, and then prevent you from, I think the use case was using adult mode in chat GPT. Do you trust those platforms with that sort of those guardrails or no?

A I think the, the guardrails could work, and it's good that there are some guardrails in place, but ultimately the business model of these platforms is to export the data of our children. This is what they are all about, and In some sense, the guardrails are running counter to their business interest. And we have seen historically that when there's this conflict of interest between what is good for the business and what is good for society, we know the choice that Zuckerberg is going to make. We know the choice that Google has made in these things. So I wouldn't really trust the guardrails put in place by a company that doesn't really have the incentive to put in guardrails.

AI assessment note: “I wouldn't really trust the guardrails put in place by a company”