The Exchanges

Q Yeah, what could be coming with the new AI regulations? Any sense as to what they might touch?

A Um, if you force me to bet, uh, I think a lot of it will focus on data security, and so it'll be kind of under the, you know, banner of AI, but it'll probably be a lot of the things we've talked about with data security, and I think with that, you know, think through the best practices of understanding, like, as a company or a service provider, like, what data you're collecting, why, where it goes, when you send it to a third party, um, One analogy a security professional gave me years ago was you should think of data, um, as sort of like toxic waste. Uh, you have it, you might need it, but you want to keep it contained, and if it starts leaking out, it's very hard to, um, kind of clean that up. Uh, and again, you know, someone said to me, said this to me years ago, but I, I think it's true, and so I think you're going to see these regulations try to, like, fence in that data much, much more so than, um, Again, we were doing 10 years ago as software builders.

AI assessment note: “I think a lot of it will focus on data security”

Q When people are making decisions to buy, how much of their decisions are predicated on AI capabilities? Because what you just told me are a number of specific, I want to build an AI program, I'm coming to Google for that. Now, I imagine that's important, but when you think about the broader landscape of people making decisions to buy cloud services, how, how much does AI factor right now?

A It's a good question. It depends on the country. It depends on the industry. It depends on the segment. Let me explain what I mean. If you're an AI unicorn, meaning you're funded to build a foundation model, or you're building an application based on AI, that's really the central part of your decision. If you're in an industry that, for example, in retail, where we have a product called, ah, retail search and conversational shopping, where you can take Google-like search using text, images, video, and put it on your catalog, And you can also put conversational shopping where I can ask a question. I'd like to return this dress and have the system handle that transaction for you. It's a super important thing, for example, for people in commerce, whether that's retail or telecommunication. On the other hand, if you look at a utility, Or an industrial manufacturer. It applies to part of their organization, but it may not be the central thing. And so it really depends by industry and by customer segment. And so, but we, part of our value proposition is that we offer all of these different capabilities. And so AI is helping us. It's not the sole reason for our growth.

AI assessment note: “And so AI is helping us. It's not the sole reason for our growth.”

Q With inference. So that will be the new cost to basically taking the models and putting them into production and using them. I'm curious how big of, um, how much of the cost of that, or how much of the use of, of your services is going to be toward reasoning and what have these new reasoning capabilities allowed your customers to do that they couldn't do previously?

A It's a really good question. I mean, reasoning is something we are starting to see customers using in different parts of our enterprise customer base. For example, in financial services, we've had people say, hey, I want to understand what's happening in financial markets, summarize the information coming off, whether that's Video feeds like CNBC, financial market indexes, and other financial information, and tell me what's the, what's happening. And the model can not only build a plan for how it collects the information, but summarize it, and then reason on the summary to say, are there, you know, conclusions to be derived, right? Uh, so we are starting to see people starting to do that. Uh, how much of that will be versus Other scenarios, time will tell, but we are starting to see people doing much more sophisticated, complicated reasoning. Even in areas, we have a travel company, for example, that's working on, give me a very high level description of what you want to travel for. I want to fly to New York. I'm taking, you know, my son. We'd like to see Coney Island and the following three things. Build me a plan. And in that, it can have multiple choices, but it may say, You know, if you're traveling in June, maybe hot in the afternoon, therefore I think we should have you see Coney Island in the morning and go to the museum in the afternoon. And models are starting to be ab…

AI assessment note: “How much of that will be versus Other scenarios, time will tell”

Q you know, it services, you really, you care about fixing stuff. You don't care about documenting it. And so this is one example, I think of how you can use gen AI sitting in the background, you go in, you remote in, you do what you need to. And then all of a sudden, all that documentation is waiting for you. Am I, am I, am I getting it right?

A Yeah, exactly. I mean, it's the toil, right? The toil that you had to deal with. It's starting to take away that toil from your day to day. And then, so think about the technician, right? The technician is now actually spending time exactly like you said, solving real problems. The toil is taken away. Uh, it's not only documenting those steps, it's creating knowledge articles. And there's a couple of additional applications from that. The first is, The next time you get a call for a similar sort of problem and a less experienced technician has to solve it, guess what? They have a head start because now those steps have been documented. They can now follow those steps. And even better, we can now actually automate and create scripts to automate those steps. So rather than have somebody go through manually click through a whole bunch of steps with Gen AI, you can take a summary and create essentially an execution plan. And start to run those. And so the world of having a human technician now actually supervising what I call virtual technicians that can execute commands on their behalf and run in parallel on a number of different machines at the same time. That world sounds a little bit like science fiction, but guess what? It's true. It's already here, right? We have our products as part of our unified endpoint management product. That product set is branded LogMeIn Resolve. We h…

AI assessment note: “Yeah, exactly. I mean, it's the toil, right?”

Q So is it really just making it more personable than the others? Like that's the way to differentiate?

A Yeah, I, I think so. I think like at the end of the day, um, we are like at the very beginning of a new era where there are going to be as many copilots or AI companions as there are people. There are going to be agents in the workplace that are doing work on our behalf. And so everyone is going to be trying to build these things. And what is going to differentiate is real attention to detail, like true attention to The personality design. I've been saying for many years now, we are actually personality engineers. We're no longer just engineering pixels. We're engineering tokens that create feelings, that create lasting, meaningful relationships. And that's why we've been obsessed with the memory, the personal adaptation, the style, and really just declaring that it is an AI companion, you know, not a tool, right? A tool is something that does exactly You know, what you intend, what you direct it to. Whereas, you know, an AI companion is going to have a much richer, more kind of emergent, dynamic, interactive style. It will change, you know, every time you interact with it, it will give a slightly different response. So I think it's going to feel quite different to past waves of technology.

AI assessment note: “Yeah, I, I think so... what is going to differentiate is... The personality design.”

Q we're definitely seeing these new modalities come out. Voice, of course, we've obviously this, we're in the middle of a firestorm with images, um, but I am, okay, I guess, let me ask the previous question a little bit differently. What, do you think that there are diminishing returns on pre-training right now? Basically scaling up the biggest possible model and then building from there? You're shaking your head no.

A Maybe specifically on pre-training, it's been a little, um, slower than it was in the previous four orders of magnitude. But the same computation, um, the same flops, or the, the units of calculation that go into turning data and compute into some insight into the model, that is just a different application of the compute. We're, we're using compute at a different stage. We're either using it at post training, or we're using it at inference time, where we generate lots of synthetic data to sample from. So net net, we're still spending as much on computation. Um, it's just that we're using it in a different part of the process, but, but for, for as far as everyone else's, you know, should be concerned aside from the technical details, we're definitely still seeing massive improvements in capabilities. And I think that's, that's for sure going to continue.

AI assessment note: “Maybe specifically on pre-training, it's been a little, um, slower”

Q Yeah, that is fascinating. Um, you know, as we're talking, I'm just thinking again about the cybersecurity. We said we'd talk about the cybersecurity side of things. Um, could you see these new agents opening up new cybersecurity challenges for companies?

A Of course. So for instance, there's a thirty-five billion dollar TAM market behind, uh, anything to do with identity. It's, it's called IAM or SIAM, so Customer Identity Access Management. And you've got companies there that are behemoths that deal with identity, but don't know how to cope with machine customers and, and AI agents because they can't decipher if a user is indeed an AI agent or not. So where does this World Cup go to? And it presents a huge opportunity for cyber criminals again around fake identities. It also, imagine you're, you were a bank robber. Imagine that was your career choice, but, but a real bank robber in the physical world, and imagine you could use a robot that looks and behaves and walks like a real human being, and that robot cannot be associated with you, or it's going to be very difficult to associate that robot with you, and you would send that robot to rob a bank on your behalf. Perfect crime. I mean, if the body's caught, How can you be associated with it? It's, it's, it's crazy. Similar to that, a lot of new, uh, criminal enterprise and criminal opportunities are gonna arise from the fact that you can use a proxy, basically, an AI agent to, to, to work on your behalf, to commit crimes on your behalf, and much, much more.

AI assessment note: “Of course. So for instance, there's a thirty-five billion dollar TAM market”

Q So how do you then work on matching the identity, uh, between, let's say, bots, customers, um, whoever's visiting a website?

A So we've done bots and non-human detection. Again, originally that was the plan also to check your AI and hence check that AI. And we've been doing it for years. Now, um, a couple of months ago, we acquired a company called deduce.com, which, uh, uh, produces what I believe is the best identity graph out there. It's certainly the best I've come across to, uh, across startups as well as incumbent players together. Marrying the identity graph together with human, not human. And obviously it's way more granular than that. It's not just human, not human. It's AI agents that are good versus AI agents that are bad and so on and so forth. Marrying those two allows us to not just understand that this is Alex's, uh, AI agent. It's also legit working on behalf of Alex from your device and so on and so forth. But that's how we're looking at it.

AI assessment note: “Marrying the identity graph together with human, not human.”

Q get to your story about, um, the future of Silicon Valley. You have a pretty provocative piece on YouTube and I want to hear your thoughts on it. Um, the title is, is Silicon Valley about to lose its monopoly on tech? My thoughts on sovereign tech stacks. So can you run us through the argument and then I'm gonna, I think I'll debate you about it a little bit.

A Yeah, yeah, real quick. Um, so, uh, this was something that I, I did a bunch of stories this week about how Europe is maybe gonna pull back on, on buying, um, US cloud infrastructure. Um, there's been a bunch of stories like that about recently, like the, uh, people don't want to buy American because they're concerned about, um, you know, here's the thing. Um, what if the rest of the world did to us what we've been doing to China in terms of like import bans, export bans, things like that, where we don't feel like your tech is safe to use internally. Um, the, you know, the, the argument that has been made that we've been covering for years now is you have to onshore, uh, chip development, silicon development, because if you, if you lose access to chips and, uh, China invades Taiwan, what, Your economy is screwed in the same way that if you lose access to oil, your economy is screwed. Um, the, the, the term that is being bandied about in recent weeks is tech sovereignty. So the argument is that if, if people felt that way about Silicon, about the chips themselves, now it's moving to the entire tech stack. If you are a country that feels like, oh, we could lose access to databases and cloud computing And even, even social media, because social media is the modern communication network in the same way that you would fear in a war that an adversary would take down your telephone ne…

AI assessment note: “So the argument is that if people felt that way about Silicon... now it's moving”

Q could do that, but it did run the code. And then also manipulating, showing, uh, uh, evaluators that it was actually modeling one behavior, whereas when it thought they weren't watching, it was doing something else. This is all stuff that's happened. Are these, are these the early signs of what could go wrong with AI, or Or is this just benign activity that we shouldn't read too much into?

A Um, so, uh, I, I think that this, it is suggestive of some loss of control scenarios. However, it is not the type of thing that I'm most concerned about with loss of control. So, that's because we still can control these AI systems somewhat reasonably right now, and maybe we'll get better methods for doing so. Um, however, the loss of control A mechanism that I'm most concerned about is when we have automated AI research and development. So imagine, at some point in the future, It could do what AI researchers do. And if you have one AI that can do that, and that would be, involve automating a lot of software engineering, if, if you could have one AI do that, then you could make, you know, a 100,000 copies of these and have them perform research simultaneously. So that could lead to some very substantial, um, acceleration in the rate of development. You might get a decade's worth of AI developments within the course of a year. And this would be highly automated, where there's very little human oversight. And I think in, in those situations or, or in that scenario, I think a loss of control is much more likely. The sort of pro, the sort of AIs that we have right now, kind of, you know, being a little nefarious here and there, um, that is a concern, but that seems more tractable as a thing to, to research and improve and, um, reduce the risks of. Meanwhile, some automated AI resea…

AI assessment note: “I think that this, it is suggestive of some loss of control scenarios.”

Q the thermostats up and down, then AI could probably do incredible amounts of very beneficial coding and computer work for humanity. So if we do get to that point, it seems to me like there's gonna be these, these maybe two poles here, right? One is the potentially scary and destructive stuff that you can mitigate, right, with some of the controls that you talked about, but also amazing opportunity.

A Yeah, so it's, it's, and the thermostat thing was for messing with the electricity and that causing strain on the power grid and, um, destroying transformers. The, just for, uh, clarification in case, uh, but, um, yeah, I think you're pointing at that it's dual use. So, um, uh, I'm not saying AI is bad in every single way, and, uh, it's, it's like other dual use technologies. Bio is a dual use technology. Can be used for bioweapons, can be used for healthcare. Um, Uh, nuclear technology is dual use. There's civilian applications for it as well. And chemicals too. And we have managed all of those other ones by selectively trying to, you know, limit some particular types of usage and restricting the capabilities of rogue actors to some of these technologies and making sure there are good safeguards for the civilian applications. Um, and then we can actually, um, capture the benefits. So it's not an all or nothing Uh, uh, type of thing with AI. Um, it's, uh, what are surgical, um, uh, restrictions one can place so that we can keep capturing the benefits. And so, for instance, with virology, that's a matter of you add the safeguards and then the researchers who want access to those can speak to sales. That's basically a resolution of that problem, provided that you have, um, the models kept, um, behind, behind APIs. And, um, Uh, so, uh, now on this dual use part, though, there's an…

AI assessment note: “I think you're pointing at that it's dual use. So, um, I'm not saying”

Q Um, What do you think about that? I mean, we just saw Deep Seek, uh, not to, you know, go back to it all the time, but it effectively equaled the cutting edge at, Uh, the proprietary labs, and, you know, put the weights on its website. So how can we possibly have a relationship of safety with AI if open source is out there exposing everything that's been done?

A So I've, um, been, I haven't been endorsing open source historically, but I've thought that releasing the weights of models didn't seem robustly good or bad. So I sort of was like, it's fine, seems to have complicated effects. There's an advantage to it, which it helped with diffusion of the technology, um, so that more people would have access to it, and sort of get a sense of, of AI, and this would increase sort of literacy on this topic, and just increase public awareness, and, um, get the world more prepared for, for more advanced versions of AI. Um, so that's been my Um, uh, historical position, but this depends on, it should always proceed by a cost-benefit analysis. So if the, if for instance they have these cyber capabilities later on, um, yeah, I think that, or I think that would be a potential place to be drawing the line on, um, on, uh, open weight releases, uh, personally. Um, Uh, in particular the ones that could cause damage to critical infrastructure. Um, you could, you could still capture the benefits by having the models be available through APIs, um, and if they're like software developers, they have access to these, you know, some more cyber offensive capabilities. But if they're a random user, they don't. If they're a random faceless user, they don't. Um, and, um, and likewise for virology. Once there's consensus, uh, once the capabilities are so high that t…

AI assessment note: “I think that would be a potential place to be drawing the line”

Q they call it, probabilistic space. So I don't know how we, we still don't have, we still haven't seen Alexa plus. Maybe it'll be come out by the time this, this interview hits. Uh, but I don't know how you do that in business, uh, because you really in a business, uh, you're kind of risking a lot if you start to leave that open. So how does that work?

A Yeah. So we, we take, um, you know, some off the shelf large language models and we actually have, um, eighteen billion interactions that are, um, that are anonymized, uh, in a database with people rating customer, uh, interactions, a positive or negative, real simple, thumbs up, thumbs down. And we do a lot of post training on our models, the basic models after the fact to try to get their probability to your point, win rate, their resolution rate as high as possible. And what's interesting, um, you know, when a, when a bot makes an error, they do make errors. It's a hallucination. When a human makes an error, you know, it's a mistake. And what we find right now is on our next generation agentic AI bot, um, we have a lower error rate than a human being, uh, from a, you know, contact center, support center, answering like for like inquiries. And so when we talk to our customers, we say, look, there are going to be some errors. A hundred percent. It is going to happen. Um, just like a human being, no matter how much you train them, how much you work with them are going to have some errors. But what we're finding generally is you can serve your customers at a, a much lower cost. You can get their, your customer satisfaction up because if you have a personalized bot that can instantaneously respond versus waiting online for five minutes or 10 minutes for someone to respond to your…

AI assessment note: “we do a lot of post training on our models... to try to get their probability”

Q Alright, so can you just walk us through, because we'd love to hear the practical examples, how would a customer basically use this platform?

A Yeah. A customer would use this platform, say, um, we, what we do is we come in, uh, for our larger customers, uh, for our smaller customers, um, we will go real quick, analyze their ticket data or their customer interaction data. And we say, we think there's an opportunity to automate 40% of those interactions. Okay. And we think we can go do that with you in two or three months or two or three weeks or two or three days, depending on the complexity. And, um, And so we're pushing some things in products to let them know what we think is going to happen. Like, we think we can go automate 40% of resolutions. We think this is going to lower your cost to serve by X. And we think it's going to take your customer satisfaction up from, depending on if you're doing CSAT, uh, top box from a four to 4.4. If you're doing MPS from a 60 to 67. And so we talk about that with them. And then what they do is we go in through, um, you know, an implementation with them in an adoption phase To go get that. We usually do an A B testing to show them, um, on the, um, on the, on the platform that we are getting what we promised. Okay. Through an A B test and they are getting the customer satisfaction. One of the biggest worries we see with our customers is I believe you can go automate X percentage of my interactions. I'm, I'm worried that customer satisfaction is going to go down and, um, companies …

AI assessment note: “what we do is we come in... analyze their ticket data or their customer interaction data”

Q if not more, we'll be, we'll be thrilled about this. And then you've brought up the co-pilot a couple of times. So What is that going to look like in a customer service agent's dashboard? And then we're going to move on to jobs because the questions about what will happen when people jobs are starting to percolate, but I want to know what it looks like in the dashboard.

A Old school human agent would get a, um, you know, information from a customer, whether it came in in a message, a WhatsApp, uh, Apple business chat, an email, a web form, you name it. Okay. And then they would craft their own reply. They'd have some canned replies on likely Interactions, uh, that they could, you know, like cut and paste basically. Okay. That's kind of old school human agent, uh, you know, resolutions. What happens now is if you have, even on a voice bot, but I'll, we'll put voice aside, any kind of digital interaction, the platform should be suggesting a response for you. So it's not you pulling as a human agent, okay, data or response into a response. It's actually, here's the suggested response. We looked at 17 similar customer interactions. This is the response that got the best, uh, response. Again, do you want to personalize it anymore? Do you want to change the tone? Do you want to accept it? Do you want to edit it? And so it's kind of changed in the human agent experience, uh, experience from a lot of times creating things on their own or, you know, pulling information in to more of an editor where they're deciding, hey, this is the right response. This, maybe it's not the right response because, um, the, um, The co-pilot does not realize how upset you are, Alex, because you had to type in agent 55 times to get to the human agent, and you've got a really…

AI assessment note: “the platform should be suggesting a response for you”

Q way about Apple before to my recollection. So to continue on our, um, on our line of question here, is this is Apple intelligence, just one bad Apple, or is it indicative like Gruber is saying of a company in disarray and not crisis? And then what do you make of the magnitude of, I would say the number one Apple watcher and perhaps fan coming out against the company?

A Yeah, he's really picking on the Apple. Um, I think it's, um, I think it's, it is really remarkable, because, I mean, Gruber has been, you know, the ultimate Apple fanboy over the years, um, and, and Apple loves him. I mean, I, I think, you know, him doing this, it's almost, it almost seems to me like a, like a spell has been lifted. He sounds like somebody who's been under a spell, I think probably for many, many years. And is now sort of waking up to reality because it's not like this is, you know, something brand new. It's not like this is a huge turn of events, right? This has been a gradual thing that's been going on for, for many years as the company has sort of scaled back on innovation, really leaned on the fact that it's, it's walled garden. Some might call it a monopoly. Um, and it really, I think used that marketing power, that goodwill that it's had, you know, since the days of Steve jobs here, To kind of show that, hey, we're not actually behind on AI. And of course they are behind on AI. And, you know, that's just the reality. And I think when you, when you come out and you sort of, and you make all these promises, I mean, it's bound to come back and bite you. And I think that was, that was obvious to anyone who was not under the Apple spell. I think immediately when they, you know, when they did that big marketing show a year ago. Uh, so, you know, it, it, it's j…

AI assessment note: “I think it is really remarkable, because, I mean, Gruber has been, you know, the ultimate Apple fanboy”

Q So how does it know when The user is saying, turn the lights on versus like something more esoteric. Like, is there something built within the technology? That's kind of like a switcher that determines first your intent, and then decides which part of the model to send it out to?

A The way to think about it is, you know, at, at the base level, you have large language models, and you have this model agnostic system that's even itself gonna choose the right model for the job, and the models play different roles in there. What, what's already happened is, um, even, even honestly, sort of in the way you asked a few of the questions, is that people assume the large language model is the product. A product like Alexa is. So much more than quote unquote, just a large language model. So you have models playing many different roles in the, in the system overall, even models helping us decide which model and models themselves deciding if they're the best, you know, tool for the job, so to speak. So then you have a system that progressively decides how to get something done. I wouldn't think about it like a switch or something in classic computer science that Is it, you know, it's a gate. That's not, that's not how the system works. It's, it's a collection of model behaviors and systems downstream of that, that complete specific tasks. And that, and that's where we introduced this term expert to try to help coalesce around the system behavior and explain it better. The large language models are interacting with these experts that do things like get you the sports score, play a song, play a video, know where you are in the song so that you can go to the video, like a…

AI assessment note: “I wouldn't think about it like a switch or something in classic computer science”

Q what's coming next, or, uh, maybe this is probably, it probably is the most impatient industry in the world and the most impatient users in the world, but Um, it seems to me like the expectations for GPT-V are built up pretty high, and so I'm curious from, like, your perspective, um, do you think it's gonna be hard to meet those expectations whenever that GPT-V model does come out?

A Well, I don't think so, and one of the fundamental reasons is because we now have two different axes on which we can scale, right? Um, so GPT-V, this is our latest scaling experiment along the axis of unsupervised learning, but there's also reasoning. Um, and when you ask about, kind of, like, uh, why there seems to be, you know, a little bit bigger of a gap in release time between four and 4.5, we've been really largely focused on developing the reasoning parallel paradigm as well. So, um, I think, you know, our research program is really an exploratory research program, right? Um, we're looking into all avenues of how we can scale our models, and over the last, you know, one and a half, two years, We've really found a new, very exciting paradigm through reasoning, which we're also scaling. Um, and, and so I think, like, uh, GPT-V really could be the culmination of a lot of these things coming together.

AI assessment note: “Well, I don't think so, and one of the fundamental reasons is”

Q For me, the future is very much aligned with niche models existing in workflows and less so of these general purpose God models. Um, so clearly open AI has a different thesis here, and I am curious to hear your perspective on what we get with the big models versus the niche models. And do you see them in competition or as compliments? Help us think about, think through that.

A Yeah. Yeah. So I think one important thing is we also serve models that are smaller, right? Like we serve our flagship frontier models, but we also serve mini models, right? Which are cost efficient ways that you can access the capabilities or fairly close to frontier capabilities for much lower cost, right? And we think that's an important part of this comprehensive portfolio here. Um, fundamentally at OpenAI though, uh, we're in the business of advancing the frontier of intelligence. And that involves developing the best models that we can. Um, and I, I think really kind of what we're motivated by is really pushing that out as much as possible. Um, we think there's always going to be use cases at the frontiers of intelligence. Um, you know, we, we think that, you know, going from 99.9 percentile in, in mathematics to the best in the world in mathematics, right? Like that difference means something to us. Like I think, uh, what You know, the best human scientists can discover is tangibly different, right, from what you or I can, can discover. So, um, we're, we're motivated by pushing the intelligence frontier as far as possible. And at the same time, uh, we want to make these capabilities cheaper and more cost effective to serve for everyone. So we don't think the niche models will go away. We want to build these foundation models and also figure out how to deliver these capab…

AI assessment note: “we don't think the niche models will go away. We want to build these foundation models”

Q Yeah, but, but again, just to hammer home on this, like, what does that improvement in model get you? Like, what do you think that it will enable?

A Yeah, yeah. So, I mean, I think, ah, I mean, agents of all forms, right? When you look at stuff like deep research, for instance, right? Um, it gives you the ability to essentially kind of Get a fully formed report on any single topic that you might be interested in, right? Um, I've used it to even put together, like, hour-long talks, um, and it goes and really kind of, like, synthesizes all the information out there and, and really organizes it, comes up with lessons, allows you to do deep discovery, um, it allows you to, uh, You know, like dig into almost any topic that you're interested in. So I feel like, um, just the amount of information and synthesis that's, that's available to you now is, is just really rapidly evolving.

AI assessment note: “I mean, agents of all forms, right? When you look at stuff like deep research”

Q candidly in the book, about what it was like to go from someone who saw people looking at you, and you're like, nah, they must not know me to now, like, you can see them out of the corner of your eyes, and you're like, oh yeah, they probably know me from the TV show. So, what has What's becoming a famous person been like, and what are the trade-offs?

A I mean, the interesting thing about fame in this era is it's incredibly relative, right, because different people have different levels of fame, so people recognize me because I have a TV show. Um, at first, there's something kind of addictive and beguiling but also strange and alienating about, like, the gaze of strangers recognizing you, and you could also see in them that they're, they're going through a strange experience because often, Our facial recognition circuitry is so profound and powerful that we recognize a face before we can place it, and so they'll have this sense where they're like, oh, Did we, did you work at Bank of America? Did we go to college together? Were we, like, they're trying to place you, and there's this familiarity that doesn't match the fact they don't know you. And then there's often this moment of like, oh, right, I don't actually know you. Like, and it's this sort of amazing moment where you're seeing the human wiring that was, like, born of 250,000 years of evolution hit up against the strangeness of modern technology. In terms of the incoming experience, I think that it can make you, it can make you vain. It can make you very aware of how other people are perceiving you. There's a level at which, like, even if you walk, if you leave the house and, like, you're unkempt, which I never would have thought of before, like, who cares? But you defin…

AI assessment note: “the loss of anonymity means that you then are viewing yourself through someone else's gaze”

Q works practically. It's like a CAI CEO is going to actually show the product, which is always cool. Before we get into that, I have to ask you a question because this is sort of stuck with me since you said it, you built a AI product basically for form, uh, completion and your customers then sort of hacked it to do customer service. I mean, how did this happen?

A It's actually like, even from day one, we had the customer service aspects, but we were thinking, okay, when people are filling out, filling out the forms, they're going to also have like questions. Like they are going to, so we have, from the beginning, we had the knowledge base aspects of the product, uh, in place. So you could upload like documents. Uh, you could give URL of your website so that we can crawl your website and learn from it. Uh, you can, uh, answer, like, questions and answers. You can just train. We even implemented this feature where you can actually talk to your own agent directly and train it, and it's asking you question, like, what if someone asks this question? You can role play with your, uh, agent.

AI assessment note: “even from day one, we had the customer service aspects”

Q so interesting that, um, you decided to then extend beyond that. To me, I, I was like, when I initially heard of your podcast, I was like, oh, Evan is sending his voice agent to scam scammers. That's great. And that's a show. But then you progress beyond that and you got really weird, um, especially sending the bot to therapy. So why did you send the bot to therapy?

A Well, it was partly because, you know, with a particular with AI, but this has happened, as you know, with many technology products over the years, the thing that the companies who are putting these products on the market will tell you is the more information you give it, the more useful it will be. So if you're going to have an AI agent, You need to give it all this information about yourself so it knows you, so it can do things for you. That is going to be the thing you're going to hear nonstop over the coming years. And so I thought, okay, well, why don't I do that? Why don't I give it all this information about myself, my mental health history, my life story, basically, and see what happens if I send it to therapists, like what, what problems will it surface? What answers will it get? And then I thought, well, first I'll send it to an AI therapist because What a perfect match. Like, my voice agent AI sitting there talking to these AI therapists, which are on the market right now. You can call them up. You can get AI therapy from an AI from a chat bot.

AI assessment note: “see what happens if I send it to therapists, like what, what problems will it surface?”

Q So talk a little bit about why people would, they just thought it was cool or what was it?

A I think some people, Thought it was cool, and, and I think they saw some humor in it initially. So I think the people who responded best, they kind of thought, This is something you're, I mean, they're used to me doing strange stories over the years, and so they might think, well, he's doing something weird, this sounds like an AI, but also, like, this must be a joke. And I think if that was the frame of mind they were in, then they, they, a couple of them loved talking to it, because of course they loved, you know, trying to egg it on to say this, that, or the other, and you can hear them, you can hear the kind of excitement in their voice, including my friend Chris, who's a lawyer, who, I sent it with actual legal questions about the show, And he answered them very succinctly. In fact, probably better than he would have answered them if I had called him myself. So it was useful in that sense. But those were the people who really kind of embraced, like, oh, I'm talking to an AI. Like, this is a new experience. I want to, I want to see this through.

AI assessment note: “I think some people, Thought it was cool, and, and I think they saw some humor”

Q could be walking around talking on like a headset, but on zoom, it will just be a, your avatar talking and looking lifelike. It really looks lifelike. And why are we not just a step away from somebody sending their AI out in work? So what do you think about that use case? And what, what should we be concerned about that? Or what, how should we feel about that?

A I feel like that use case, to me, it comes with a lot of the issues that a lot of AI products to me come with, which is that they, the, the, the people who have designed them generally have one set of problems that do not apply to most humans on the face of this earth. So yes, the Zoom CEO would not like to be in meetings. The Zoom CEO would like to send a digital twin to meetings In his stead. Great. Nobody wants to be in meetings. Most people do not want to be in meetings. So do the other people get to send theirs or just the CEO? And then the question is, if everyone sends their agents to meetings, who's going to process all the information? Like, they're going to distill it for you. Like, what's the purpose of the meeting? What is the purpose of the work? Like, I feel like those things all get lost in these discussions. And what ends up happening is super busy CEOs and very wealthy people come up with solutions for them, and you kind of wonder, like, well, what happens with the rest of us? And so, I feel like that stuff is right around the corner. I know people who have gone to job interviews, to meetings, where they encounter an AI when they do not expect to encounter an AI, and I think we're gonna see more and more of that in the next months, years.

AI assessment note: “I feel like that use case, to me, it comes with a lot of the issues”

Q are you on the question of AI sentience? I mean, you basically created an AI to resemble a person. I know we don't think, or at least I don't think you I believe that AIs are sentient now, but do you, did you feel any, like, say in AI world, uh, feel the AGI at all, or did you feel any hints of, um, personhood and AI that you developed?

A I did not. In fact, the more that I deployed it, the less I felt that. Now, I, I think when we're talking about how close is AGI and those types of questions, I think the people inside the companies Who are extremely closed about what's going on. Um, they're dealing with non-guard railed versions of the chatbots. So I think they may have different experiences and they write about, you know, as you talked about in a recent show, like, the chatbot's lying and things like that. In this case, you've got the fully guard railed, you know, chat GPT, latest version. The more that you talk to it, the more generic it feels to you. The more you can actually feel The training data, the, the sort of, like, distilling down the training data and the predictive aspect. Like, oh, it's trying to predict what a human would say in this given moment, and what the average human would say in this given moment is actually quite lame. Like, that's what you really figure out. So, I felt further away from that the more that I, like, spent time with it. Now, I don't think that's necessarily, like, a statement about how close or far away it is, because I think those aspects are all internal to these companies, and, like, we just don't have access to them.

AI assessment note: “I did not. In fact, the more that I deployed it, the less I felt that.”

Q So why don't you take us down the road a little bit. I mean, what do you think is going to happen as this technology gets better?

A I think, uh, first of all, as with many things, the market is going to dictate that people are going to use this technology, even if it remains flawed, even if it remains not quite human quality, uh, for all sorts of, I mean, the very obvious ones are like telemarketing, call centers, uh, ordering food at a drive through, you know, places where they can save a little bit of money by deploying it, even if it messes up sometimes, even if it does crazy stuff, like give you the wrong order, like, They'll just say, well, humans mess up too, and it messes up less. And so I think we're going to start to see them infiltrate these different parts of society. And I think the question is going to be how people respond to them. And if people are sort of like, eh, it's same, same to me, or maybe this customer service AI is actually more helpful than the person that I get sometimes when I call the social security administration or the VA or whatever benefits I need, maybe people will embrace them. I think you have seen some instances with technology where like, With, you know, for instance, check out, checking yourself out of the grocery store where a lot of people don't like it, and then maybe they go back to humans. So I think it's still, the balance still is yet to be determined, but I think there is no question that voice AI agents are just going to be, they're just going to be deployed …

AI assessment note: “we're going to start to see them infiltrate these different parts of society.”

Q And so who's going to use it? Is it going to be robotics developers? Is it going to be somebody that's building, let's say LLM based application, but just wants them to be a little smarter? Both?

A It will be all of them. Yes. It's, uh, we, We feel that we're as a, as, as the industry, the world is right at the beginnings of this physical AI revolution, and no one company, no one organization is going to be able to build everything that's, that, that we need. So, so we're building it out there in the open, uh, to encourage others to come build on top of what we've built and come build it with us. And this is gonna be, ah, essentially anybody that has an application that involves the physical world. And so that's definitely robotics companies are part of this and, and robotics in the very general sense that includes self-driving car companies, robo taxi companies, and as well as, uh, companies building robots that are in our factories and warehouses. Anybody that wants to make intelligent robots that have perception and operate autonomously inside the real world, they want this, but, um, it's not, It's not only about robots in the way we think about them as, as these agents that move around. Uh, we have sensors that we're placing in our spaces, in, in, in our cities, in urban environments, inside buildings. Uh, these sensors, uh, need to understand what's happening in that world. Maybe for security reasons, for, for coordinating, um, other robots, changing the climate and, And, uh, energy efficiency of, of our buildings and data centers. So there's, there's many applicatio…

AI assessment note: “It will be all of them. Yes.”

Q of comments. Uh, it seems to indicate to me that China would, so for folks who are listening, who don't know about it, it is a large language model out of China, um, according to its own benchmark. So Take this with, you know, for what it's worth. It outperforms Meta's Llama 3.1 and open AI's GPT four. Oh, um, it seems like China's in the game. Do you agree?

A Uh, absolutely. Uh, and it was one of the conversations that I'd frequently have the last couple of years where they'd say, oh, you're just bringing up China competition as kind of the bug bear. We're like, no, no. Uh, you know, part of my job as a investor, as an entrepreneur, as a theorist is to predict where they, Puck is moving to, and the Chinese are strong and vigorous. There's lots of very capable, very hardworking entrepreneurial folks with technological depth, um, and, and so it doesn't surprise me that DeepSeq has, has, has entered the field, um, with, with vigor, and I think that, you know, part of when you look at, um, kind of what is going to be playing out with AI, um, China and multiple companies within it, because, you know, I haven't evaluated yet. Minimax, which also just came out recently, but will have a number of very strong contenders, uh, for this, uh, in the, the world technology landscape. And so I think it's a, um, you know, the, the game is on or game is afoot if we want to be, uh, Sherlock Holmes.

AI assessment note: “Uh, absolutely.”

Q I think it's always nice because, you know, the things that people fire off in a, in a moment, maybe that's pretty telling in terms of what they think about the world. So a couple from you, uh, one, very soon voice will become the main interface, And AI will get better at learning what exactly you're looking for. So you really think voice is going to be, why voice?

A A hundred percent. One of the party tricks I do is I bring out my phone and put ChatGPT on audio mode to show people. Because part of when we learn to use phones or PCs, we've learned, you know, with GUIs and everything else, we've made it softer, but there's a very precise semantics. One of the things that, that language models allow us to do is to talk to it. And so now, like, you can prompt, you know, ChatGBT with just like, kind of almost like word salad and still get something very interesting. So you don't have to, and, and so it allows you to be more, you know, kind of inchoate and, and, and, and speculative and brainstorming with it. And you can actually have a useful conversation and create useful artifacts from that, whether it's on your phone or PC. And so, um, that's part of the reason why I've, I've, I've, you know, made that as a prediction.

AI assessment note: “One of the things that, that language models allow us to do is to talk”