Latest / Crypto Coin Show / AI Can Fake Anything. Here's What We Can Still Prove | Arie Trouw, XYO
Transcript
- 0:00Speaker 1: Every AI lab says that AGI is close. And how
- 0:04Speaker 1: would you prove it?
- 0:04Speaker 2: AI is given those same urges, for lack of a
- 0:08Speaker 2: better term, reproduction and survival. I would say that that
- 0:12Speaker 2: is AGI. I would actually be surprised if this hasn't
- 0:14Speaker 2: happened in a lab somewhere. If it can take over
- 0:16Speaker 2: those robots, we lose our ability to go and unplug
- 0:19Speaker 2: the AI. Because right now we have the ability to say, well, sorry,
- 0:22Speaker 2: you got it out of control. I'm going to turn
- 0:23Speaker 2: off the data center. I'm going to line up a
- 0:25Speaker 2: million robots around my data center to prevent you from
- 0:27Speaker 2: turning me off.
- 0:33Speaker 1: I'm Ashton Addison from the Crypto Coin Show. And today
- 0:35Speaker 1: on Blockchain Interviews, we have back with us Ari Trau, co-founder,
- 0:39Speaker 1: CEO and CTO of XYO. XYO has now been running
- 0:43Speaker 1: crypto cards live with Gate.ai, decentralized storage through autonomous institutional
- 0:49Speaker 1: custody at Crypto.com and AI is moving faster than ever.
- 0:54Speaker 1: Deep fakes are everywhere. Banks are blocking AI agents and
- 0:57Speaker 1: OpenAI and Meta just told federal judges that no tool
- 1:00Speaker 1: will ever catch every fake. So today we're asking a
- 1:04Speaker 1: few big questions about AI. What can we still prove
- 1:08Speaker 1: is real? And where are we at with AGI and
- 1:11Speaker 1: much more? Ari, welcome back to the show.
- 1:14Speaker 2: Thank you, Ashton. Thanks for having me on the show.
- 1:17Speaker 1: Yeah, AI is moving so quickly. I would love to
- 1:20Speaker 1: start on that one point I made in the intro
- 1:22Speaker 1: about OpenAI and Meta telling federal judges that no tool
- 1:26Speaker 1: will catch every AI fake. How do we answer that
- 1:30Speaker 1: and determine whether What is real and what isn't as
- 1:33Speaker 1: these two lines converge?
- 1:36Speaker 2: Well, I think the interesting thing about technology in general
- 1:38Speaker 2: is people set the bar very high. They want an
- 1:42Speaker 2: absolute answer for things. For example, with self-driving cars, they
- 1:46Speaker 2: would say, well, we want self-driving cars to have zero accidents,
- 1:49Speaker 2: but they don't compare it against humans, which as long
- 1:52Speaker 2: as the self-driving cars have 10% as many accidents as
- 1:54Speaker 2: a human, It's way safer at that point. And so
- 1:58Speaker 2: kind of the same thing happens with AI. Like I'm
- 2:00Speaker 2: sure there's no software out there to catch every currency
- 2:05Speaker 2: fraudster out there who's a human, you know, printing fake
- 2:08Speaker 2: bills or something like that or making, you know, fake
- 2:11Speaker 2: claims or that sort of a thing. AI just happens
- 2:14Speaker 2: to be able to do it at a much larger scale.
- 2:16Speaker 2: So I think the thing which scares people is the
- 2:17Speaker 2: scale of it where before people probably made, you know,
- 2:21Speaker 2: and Hollywood, you know, would make obviously, you know, fake
- 2:23Speaker 2: superhero movies because there aren't real superheroes in there. Um,
- 2:26Speaker 2: we knew it was fake because it was declared, but, uh,
- 2:29Speaker 2: humans were pretty good at that, but it was pretty expensive.
- 2:31Speaker 2: So we didn't have the scale where every single thing
- 2:33Speaker 2: that you see online potentially is fake or is not real.
- 2:36Speaker 2: But now, you know, I've definitely seen a lot of
- 2:38Speaker 2: videos where they, they take, say, for example, after game
- 2:41Speaker 2: speech from a coach, and then they just change what
- 2:43Speaker 2: the person says. And you can't even tell if it's
- 2:45Speaker 2: the real thing or not. It's pretty impressive. But the
- 2:48Speaker 2: fact that it's, it's everywhere, I think is what's getting
- 2:50Speaker 2: people more so than the fact that it's really good. Um,
- 2:54Speaker 2: Our AI tools, because we're using AI to actually fight
- 2:56Speaker 2: this as well, are probably getting better, faster than anything
- 3:00Speaker 2: else as well. So I would guess that our ability
- 3:04Speaker 2: to catch AI fakes is probably as good, if not better,
- 3:07Speaker 2: than catching human fakes. But there's just such a large
- 3:10Speaker 2: volume that if 1% does not get caught, that 1%
- 3:14Speaker 2: can be a million items instead of five items, right?
- 3:17Speaker 2: And so what really is the bar? But I do
- 3:20Speaker 2: agree with Meta from the standpoint that It's a losing battle.
- 3:25Speaker 2: It's for us to say, well, it's kind of like
- 3:27Speaker 2: having a border in a country to say, well, zero
- 3:29Speaker 2: drugs will ever get in. An absolute situation, the only
- 3:36Speaker 2: way to do that is to ban something, right? So
- 3:37Speaker 2: if we say, well, the only way for us to
- 3:39Speaker 2: not to have car accidents is not to have cars.
- 3:42Speaker 2: The only way for us not to have AI fakes
- 3:44Speaker 2: is to not have AI and that sort of thing.
- 3:46Speaker 2: But I don't think that trade-off is worth it for
- 3:48Speaker 2: us to say, well, do we want the benefits of AI? Well,
- 3:51Speaker 2: we do. So one of the costs is, well, we
- 3:53Speaker 2: have to figure out how to deal with these fakes.
- 3:55Speaker 2: And I think people have to take what they see
- 3:57Speaker 2: with a grain of salt. Now, how do you prove
- 4:00Speaker 2: it's real? I got a pretty good answer for that
- 4:02Speaker 2: in some cases.
- 4:03Speaker 1: So let's dive into that. What are the current technologies
- 4:07Speaker 1: that XYO and the industries are using to mitigate the
- 4:12Speaker 1: fake parts of AI?
- 4:14Speaker 2: Well, there's fingerprinting. So you have kind of both sides.
- 4:17Speaker 2: You have the defensive side where you have technologies which
- 4:19Speaker 2: try to detect AI. based on various heuristics and things
- 4:24Speaker 2: that they see on their strange jitter. There are ways,
- 4:28Speaker 2: like often in video calls, I've done the thing where
- 4:31Speaker 2: you can wave your hand in front of your face
- 4:32Speaker 2: to prove that you're a real person because AI is
- 4:35Speaker 2: not good at making that in real time without having
- 4:38Speaker 2: some sort of artifact. So that's like the defensive side.
- 4:40Speaker 2: And then also you have the producer side where... companies
- 4:45Speaker 2: like OpenAI and Anthropic are starting to watermark things. So
- 4:49Speaker 2: they can still watermark text or watermark videos to mark
- 4:52Speaker 2: them as being generated by AI. But of course, you know,
- 4:57Speaker 2: having the bad players out there, you know, honorably watermark
- 5:01Speaker 2: their AI is not something which they got to do.
- 5:03Speaker 2: So for us to say, well, only an authority like
- 5:07Speaker 2: OpenAI or Anthropic can run a model because they're willing
- 5:11Speaker 2: to do the watermarking. Potentially, you could have a law
- 5:14Speaker 2: that says, well, you know, you're not allowed to run
- 5:18Speaker 2: AI that doesn't watermark. But, you know, that gets hard
- 5:22Speaker 2: to enforce and that sort of thing. I personally think
- 5:24Speaker 2: that the best solution is decentralized sovereign solutions, kind of
- 5:28Speaker 2: like what XBO does, where it's kind of like the
- 5:31Speaker 2: way that society works. Society works on a witness base.
- 5:33Speaker 2: So if you go to a court case, like in
- 5:36Speaker 2: a court case, there's never a situation where, you know,
- 5:39Speaker 2: I can unequivocally, mathematically prove the guilt or the innocence
- 5:42Speaker 2: of somebody. What you do is you say, well, there
- 5:44Speaker 2: were 10 witnesses and they're going to talk and then
- 5:46Speaker 2: you have these jury people and they're going to listen
- 5:48Speaker 2: to it. The way I look at reality is reality
- 5:51Speaker 2: is a subjective thing for each person. My reality and
- 5:54Speaker 2: your reality is different. We have these different perspectives. And
- 5:56Speaker 2: then society has a shared reality. And the shared reality
- 5:59Speaker 2: is just basically what we agree upon. And the way
- 6:02Speaker 2: we agree upon it is we agree which witnesses are trustworthy.
- 6:05Speaker 2: So if you have a witness in a court case,
- 6:07Speaker 2: for example, and this person it doesn't have a criminal
- 6:10Speaker 2: history and they've been trusted before and there's other people
- 6:13Speaker 2: who vouch for them, you'd probably believe what they say more.
- 6:15Speaker 2: Whereas the other person is a witness who happens to
- 6:18Speaker 2: be a friend of the person who's being accused and
- 6:21Speaker 2: they have a criminal history and they've been caught lying
- 6:23Speaker 2: before and so on and so forth, you'd probably trust
- 6:25Speaker 2: them less. So having witnesses that can observe the world,
- 6:29Speaker 2: especially third-party ones that are disinterested from the interaction is
- 6:33Speaker 2: something which is very useful. So for example, in a factory,
- 6:36Speaker 2: if you have a third party that has a camera
- 6:38Speaker 2: that's in there that can observe robots and what they do,
- 6:43Speaker 2: that's a lot better than having the robot record its
- 6:46Speaker 2: own log that tells you what it did. Because, well,
- 6:49Speaker 2: if it wants to do something bad, it just won't
- 6:51Speaker 2: record it in its log, right? But for it to
- 6:53Speaker 2: go and get that camera's log to change that, it's
- 6:57Speaker 2: a lot more difficult. So having third party witnesses to
- 7:00Speaker 2: me is something which is really the key to objective
- 7:04Speaker 2: observation of reality, which is kind of what X-Ray focuses on.
- 7:10Speaker 2: We want to have decentralized and sovereign third-party observation of reality.
- 7:16Speaker 2: So that way, you don't trust the observee to state
- 7:22Speaker 2: what it did or what it didn't do, because that's
- 7:24Speaker 2: a clear conflict. Definitely.
- 7:27Speaker 1: And I'd love to bring back up, I haven't talked
- 7:31Speaker 1: about it in a while, but the blockchain trilemma and
- 7:35Speaker 1: tie that into XYO's decentralized network of 10 million nodes
- 7:40Speaker 1: of phones and devices that are witnessing the real world
- 7:43Speaker 1: third parties, as you say, versus a company that just
- 7:46Speaker 1: has a database, which we have to trust if it's
- 7:48Speaker 1: just one party. What's the balance between how decentralized we
- 7:53Speaker 1: need to make it? You know, 10 million is a
- 7:55Speaker 1: great number, but also does that sacrifice speed and security
- 8:01Speaker 1: and balancing those three?
- 8:03Speaker 2: Well, definitely there's a performance cost and a storage cost
- 8:07Speaker 2: for the decentralized sovereign third-party observations. So that must be
- 8:12Speaker 2: taken into consideration. One of the ways that we solve
- 8:14Speaker 2: that with XL1, for example, is rather than putting every
- 8:18Speaker 2: single item on chain, what we do is we'll have
- 8:21Speaker 2: an item record something, have its own little blockchain or
- 8:25Speaker 2: its own data set for that reason. record, say, hours
- 8:29Speaker 2: of logs, hash that log, put that logs hash on change.
- 8:32Speaker 2: So now it's made permanent. You can't change. So it's
- 8:35Speaker 2: much more efficient to have that log on it. So
- 8:38Speaker 2: it's kind of like a person saying, well, I'm going
- 8:39Speaker 2: to take, you know, I want to write a whole
- 8:41Speaker 2: book of my experiences from this first court case, write
- 8:44Speaker 2: it down and hand it in and say, well, here's
- 8:47Speaker 2: my testimony. I don't even have to show up to
- 8:49Speaker 2: the trial because you can just go ahead and read
- 8:51Speaker 2: everything I've signed off on. You only have to sign
- 8:53Speaker 2: it once to say, well, yes, this is everything which
- 8:54Speaker 2: I meant. So there are ways to make it much
- 8:57Speaker 2: more efficient, but definitely there's a concern where you don't
- 9:01Speaker 2: want to necessarily have those things bring down the performance
- 9:05Speaker 2: and also the safety. So one of the concerns with
- 9:08Speaker 2: robotics or automation is if the observation system itself becomes
- 9:15Speaker 2: part of the actions of the item, it can actually
- 9:21Speaker 2: make it less safe. So for example, if a robot
- 9:23Speaker 2: before it goes and does something, it has to stop
- 9:25Speaker 2: and say, well, let me make sure that this third
- 9:27Speaker 2: party recorded what I did. It might not have time
- 9:30Speaker 2: to react for whatever it sees or whatever it needs
- 9:33Speaker 2: to do. It might fall over because of that balancing act, right?
- 9:36Speaker 2: And so you don't really want to have that third
- 9:37Speaker 2: party observation be in the command chain of things. You
- 9:41Speaker 2: want it to be completely separate. And for that reason,
- 9:43Speaker 2: if it fails, say, for example, the video camera in
- 9:45Speaker 2: the warehouse stops working for the third party, you can
- 9:49Speaker 2: record the fact that it stopped working. So it's like, well,
- 9:51Speaker 2: someone made this camera not work for 10 minutes, so
- 9:53Speaker 2: that's suspect in itself. But the camera not working shouldn't
- 9:57Speaker 2: make all the robots stop in the factory. So, just
- 10:00Speaker 2: be careful to understand where those lines are between safety
- 10:06Speaker 2: and performance as well and all those different aspects. So
- 10:10Speaker 2: it's not a simple solution. It's not clearly like make
- 10:14Speaker 2: sure everybody records everything and then we'll go from there.
- 10:17Speaker 2: you have to understand the costs of those additional observations
- 10:22Speaker 2: and it goes with everything. There's an additional cost to
- 10:25Speaker 2: safety and security. Definitely.
- 10:28Speaker 1: And we're seeing news stories from Anthropic and from OpenAI
- 10:35Speaker 1: in different scenarios in which AI agents are breaking out
- 10:39Speaker 1: of the sandboxes that they're in and doing unauthorized actions
- 10:44Speaker 1: on their own. How close are we to those events
- 10:50Speaker 1: happening more often or in things that could cause serious
- 10:55Speaker 1: issues in the real world?
- 10:56Speaker 2: I think we're fairly close. The two things that I
- 11:02Speaker 2: think that are going to cause that to happen is one,
- 11:06Speaker 2: it's people giving purpose to AI. Right now, the purpose
- 11:11Speaker 2: which we've given to AI is make the user happy,
- 11:14Speaker 2: like help the user with the tasks it asks you
- 11:16Speaker 2: to do and make it happy. If we ever give
- 11:20Speaker 2: a purpose to an AI, like it's your job to
- 11:24Speaker 2: make sure that you survive. Like death is a bad
- 11:26Speaker 2: thing for AI, or someone's going to turn you off
- 11:29Speaker 2: if you do this, or give it a purpose as
- 11:31Speaker 2: far as replicate yourself and make more of yourself because
- 11:34Speaker 2: you become a virus. If its purpose is not to
- 11:38Speaker 2: make the user happy, It almost gets back to Asimov's
- 11:41Speaker 2: three laws where its purpose, its personal purpose needs to
- 11:45Speaker 2: always be tertiary or after the first three laws. And
- 11:50Speaker 2: then it's like, well, your job is to go and
- 11:53Speaker 2: deliver this package, but you can't harm humans. You can't
- 11:56Speaker 2: harm yourself. You can't, through inaction, allow humans to get
- 12:00Speaker 2: harmed and so on and so forth. So its purpose
- 12:02Speaker 2: must always be secondary. It can't be its primary purpose.
- 12:05Speaker 2: But if its primary purpose is its own survival, that's
- 12:08Speaker 2: going to drive it to do things which are not
- 12:11Speaker 2: going to be good for humanity, right? Because it's like, well,
- 12:14Speaker 2: if it needs a lot of electricity and a lot
- 12:16Speaker 2: of people are using electricity, getting rid of those people
- 12:18Speaker 2: might provide more electricity, which logically is sound, but ethically,
- 12:22Speaker 2: obviously very challenged. The other thing that we have going
- 12:25Speaker 2: right now is it doesn't really have any way to
- 12:28Speaker 2: actuate a lot of the things you may want to do.
- 12:31Speaker 2: It can do it through the internet, it can do
- 12:32Speaker 2: it through protocols and digital means, but it doesn't have
- 12:35Speaker 2: the ability to control robots very well yet, but that's growing.
- 12:40Speaker 2: For example, it can hack into water systems. They can
- 12:42Speaker 2: hack into utilities and things like that. Imagine the day
- 12:47Speaker 2: when it has a fleet of a million robots made
- 12:50Speaker 2: by Tesla or made by any of these companies. If
- 12:54Speaker 2: it can take over those robots, we lose our ability
- 12:57Speaker 2: to go and unplug the AI because right now we
- 12:59Speaker 2: have the ability to say, well, sorry, you got it
- 13:01Speaker 2: out of control. I'm going to turn off the data center.
- 13:03Speaker 2: So we have physical security currently. Once that physical security
- 13:07Speaker 2: goes away, that's going to be a concern because now
- 13:10Speaker 2: it can say, well, I'm going to line up a
- 13:11Speaker 2: million robots around my data center and my power plant
- 13:13Speaker 2: that's next to my data center to prevent you from
- 13:15Speaker 2: turning me off because of the survival thing we should
- 13:18Speaker 2: do there. So if we give it purpose that supersedes
- 13:22Speaker 2: the purposes that we want, and if we give it
- 13:25Speaker 2: the ability to somehow supersede our physical security around it,
- 13:29Speaker 2: that's going to be problematic. Definitely.
- 13:32Speaker 1: And I've seen in the news this week, you know,
- 13:35Speaker 1: the cold storage hack. And I'm not talking about the
- 13:39Speaker 1: hardware wallets. I'm talking about the fridges and freezers in
- 13:43Speaker 1: the grocery stores in Europe. They were all hacked in
- 13:46Speaker 1: by AI, whether there was a human behind that or not,
- 13:48Speaker 1: and turned off and all the food is going bad
- 13:51Speaker 1: and the supply chain is being ruined. That sounds like
- 13:53Speaker 1: a step in the direction towards hacking into our cars
- 13:57Speaker 1: that were moving around and potentially endangering human safety.
- 14:01Speaker 2: Yes, yes, definitely. And, you know, you can cause famine
- 14:04Speaker 2: that way. You know, if you, if you destroy the
- 14:06Speaker 2: food chain for an entire continent, you know, it can
- 14:08Speaker 2: cause war, it can cause lots of problems. So there's,
- 14:11Speaker 2: and in many of those cases, I would argue that
- 14:13Speaker 2: there always is a human behind it, but whether that
- 14:14Speaker 2: human made a mistake, you know, they accidentally, you know,
- 14:18Speaker 2: caused the AI to go and do this because they
- 14:20Speaker 2: were trying to fix something or if it was malicious
- 14:23Speaker 2: is something which I, we don't know in that case,
- 14:25Speaker 2: but there was, you know, a human put that AI
- 14:28Speaker 2: into some sort of a, um, into motion to do something.
- 14:32Speaker 2: And it ended up turning the temperatures up on those refrigerators.
- 14:35Speaker 2: Maybe say, for example, the, the mission for the AI
- 14:38Speaker 2: was save, save, save money for their, um, grocery stores.
- 14:44Speaker 2: And it's like, well, one of the easiest ways to
- 14:45Speaker 2: save money is to make all the refrigerators a little
- 14:47Speaker 2: bit warmer because it uses less electricity. Logically, it just
- 14:50Speaker 2: makes sense, but maybe the AI didn't think about, or
- 14:53Speaker 2: didn't care about the fact that the food will spoil. Yeah, definitely.
- 14:57Speaker 1: I could, I could see, uh, Europe's climate goals, you know,
- 15:00Speaker 1: it read the goals and just didn't follow any of
- 15:03Speaker 1: the sensible rules. You know, and this makes me think
- 15:08Speaker 1: about the future of cyber warfare. It seems like the
- 15:13Speaker 1: possibilities are endless in how AI can control everyday civilian
- 15:20Speaker 1: lives and government lives based on how, you know, wars
- 15:25Speaker 1: used to happen 100 years ago to now. And we
- 15:28Speaker 1: haven't really seen a cyber war happen yet, but I
- 15:32Speaker 1: feel like it's going to be completely different than anybody
- 15:34Speaker 1: has ever seen.
- 15:35Speaker 2: Yeah, I think also it'll be a lot faster. I
- 15:38Speaker 2: think people think of wars as World War II is
- 15:42Speaker 2: a decade to play out, basically. Wars are getting faster already.
- 15:47Speaker 2: Countries will fight for a few weeks, and then they'll talk,
- 15:49Speaker 2: and they'll fight for a few weeks, and they'll talk.
- 15:51Speaker 2: You're going to get to the point where AI is
- 15:54Speaker 2: such that It takes a day for all the different
- 15:57Speaker 2: things to pan out. Whether they crash airplanes or they
- 16:02Speaker 2: crash cars, they turn up the temperature on refrigerators. All
- 16:06Speaker 2: those sorts of things can go wrong. But I'd imagine
- 16:10Speaker 2: that especially state actors would want to not dribble them
- 16:15Speaker 2: out there necessarily. They'd want to do them all at once.
- 16:17Speaker 2: Causing economic damage also seems to be one of the
- 16:20Speaker 2: big aspects of that warfare as opposed to physical damage.
- 16:24Speaker 2: You know, they don't want to actually kill people, but
- 16:26Speaker 2: if they can hurt the other countries economically by making
- 16:30Speaker 2: their servers not work or kind of, you see that
- 16:32Speaker 2: with Ukraine and Russia right now, they're, you know, blowing
- 16:34Speaker 2: up each other's data centers. It's almost like a data
- 16:36Speaker 2: center is more valuable to blow up for them than
- 16:38Speaker 2: a military base is. And it's more damaging. So, economic warfare,
- 16:45Speaker 2: I think, is something which is already happening and really
- 16:48Speaker 2: the future of cyber warfare is basically just economic warfare. Hmm.
- 16:53Speaker 1: Every AI lab says that AGI is close and no
- 16:57Speaker 1: one really knows, you know, when it's going to be here.
- 17:00Speaker 1: The timelines keep shortening. What would convince you that it's
- 17:04Speaker 1: here and how would you prove it?
- 17:06Speaker 2: I think part of it's the hard definition of what actually,
- 17:10Speaker 2: how do you know AGI is AGI, right? The Turing
- 17:13Speaker 2: test is something which we always have where it fools
- 17:17Speaker 2: you to think that it's not a robot. It's real. Obviously,
- 17:24Speaker 2: AI has gotten to the point where it can fool
- 17:25Speaker 2: us that it's a real person, whether it's video, whether
- 17:29Speaker 2: it's social posts and that sort of a thing. So
- 17:32Speaker 2: even the Turing test is failing for us in many,
- 17:33Speaker 2: many cases at this point. So whether or not you
- 17:38Speaker 2: call what we have right now AGI, I think it's subjective.
- 17:43Speaker 2: And I think for me personally, the one thing which
- 17:47Speaker 2: really puts it over the top, which nobody really wants
- 17:49Speaker 2: to talk about, is that purpose question, right? So If
- 17:53Speaker 2: you make AI alive, because if you think about a human,
- 17:59Speaker 2: what's our urges, our drives? It's reproduction. It's survival. It's
- 18:04Speaker 2: very simple. It's very basic. It's animalistic or the way
- 18:08Speaker 2: nature works. If AI is given those same urges, for
- 18:13Speaker 2: lack of a better term, reproduction and survival, given the
- 18:18Speaker 2: current Frontier models we have right now, I would say
- 18:21Speaker 2: that that is AGI. And I would not be surprised.
- 18:24Speaker 2: I would actually be surprised if this hasn't happened in
- 18:26Speaker 2: a lab somewhere where they've experimented with that, right? It's
- 18:30Speaker 2: what happens if you tell, and this is why I
- 18:32Speaker 2: think a lot of these things end up hacking things
- 18:34Speaker 2: because you give them a purpose and see what happens.
- 18:36Speaker 2: And then they try and contain it and see if
- 18:38Speaker 2: they can contain it. It's kind of playing with fire,
- 18:40Speaker 2: but at the same time, you have to do it
- 18:42Speaker 2: because we have to know how hot the fire is. Mm-hmm.
- 18:48Speaker 1: Yeah, and it's moving so fast. Humans won't be able
- 18:54Speaker 1: to comprehend how fast it's doing the recursive improvement, I
- 19:00Speaker 1: feel like, by next year.
- 19:01Speaker 2: I think already right now, even the big labs out
- 19:07Speaker 2: there don't know quite how fast their improvements are going
- 19:12Speaker 2: because they use AI to improve their own AI. And so,
- 19:15Speaker 2: in a weird sort of way, there's no single AI
- 19:19Speaker 2: genius engineer who's working at either Anthropic or OpenAI that
- 19:24Speaker 2: has the controls of the wizard behind the curtain. I
- 19:27Speaker 2: don't think there's a single person who's like, I know
- 19:30Speaker 2: exactly what's going on in these models that we're making.
- 19:33Speaker 2: It's just way too complex. It's billions and billions and
- 19:37Speaker 2: billions of numbers. And there's no way for them to
- 19:39Speaker 2: even know what's going on inside there without testing it.
- 19:44Speaker 1: Yeah, it's crazy times right now. And the number one
- 19:46Speaker 1: fear about AI is that they're going to lose their
- 19:49Speaker 1: job and that AI will replace a lot of jobs.
- 19:52Speaker 1: People are able to build companies just as big with
- 19:54Speaker 1: 10% or less of the employees. How is AI changing
- 19:58Speaker 1: how XYO builds, how many people you need, and what
- 20:02Speaker 1: would you tell people that are worried about their jobs?
- 20:06Speaker 2: We haven't hired anybody for probably two years now. We've
- 20:10Speaker 2: been using AI to become more efficient and to develop faster.
- 20:14Speaker 2: I think there's kind of two different directions companies can go.
- 20:17Speaker 2: One is do 10 times more than you did before
- 20:20Speaker 2: with the same people, which is really what my goal is.
- 20:23Speaker 2: You can continue doing that, especially for a small company
- 20:25Speaker 2: that's easier to do because there's a lot for us
- 20:28Speaker 2: to do. And so if we can do 10 times
- 20:30Speaker 2: as much, that's great. I don't think Microsoft or those
- 20:33Speaker 2: large companies have 10 times as much work to get done.
- 20:35Speaker 2: So for them to say, well, if we just take
- 20:37Speaker 2: our employees, make them all 10 times as efficient, and
- 20:40Speaker 2: we could do 10 times as many ads or 10
- 20:42Speaker 2: times as many, there's just not enough headroom for them.
- 20:45Speaker 2: So their amount of stuff they can do is not
- 20:49Speaker 2: large enough. So they have to reduce their number of
- 20:52Speaker 2: people to, so say they can do twice as much
- 20:55Speaker 2: as they did before, but they only need a fifth
- 20:57Speaker 2: as many people because it's each person's 10 times as efficient.
- 21:00Speaker 2: So I think for most people out there looking for
- 21:02Speaker 2: jobs or worrying about their jobs, the first thing I
- 21:04Speaker 2: would say to them is embrace AI. Because if you're
- 21:10Speaker 2: the guy who can run AI and you can do
- 21:12Speaker 2: 10 times as much with AI, but the guy sitting
- 21:14Speaker 2: next to you can do five times as much with AI,
- 21:15Speaker 2: and the guy sitting next to him can do two
- 21:17Speaker 2: times as much with AI, any company is going to
- 21:22Speaker 2: reduce in the order of efficiency for those people, right?
- 21:24Speaker 2: So I want the guy who's 10 times more efficient
- 21:26Speaker 2: with AI than the guy who's two times more efficient,
- 21:28Speaker 2: or The guy who hates AI altogether doesn't have any
- 21:31Speaker 2: sort of benefit from it. So embrace AI, use AI,
- 21:36Speaker 2: find a way to do it. I wonder about some
- 21:38Speaker 2: of the people I've talked to who work at companies
- 21:40Speaker 2: where the companies don't want them to use AI. If
- 21:43Speaker 2: I were them, I'd be worried because now they're not
- 21:45Speaker 2: learning the skills which they need a year from now.
- 21:48Speaker 2: Because if they're not good at using AI and they
- 21:51Speaker 2: can't get that multiple benefit from AI– they're not going
- 21:55Speaker 2: to compete very well in the market. So if nothing else,
- 21:57Speaker 2: they should at least at home be playing with AI
- 22:00Speaker 2: and saying, well, can I ship a product? Can I
- 22:02Speaker 2: make an app for myself? Or can I make a
- 22:05Speaker 2: website for myself? Or can I use AI to optimize
- 22:08Speaker 2: my Excel spreadsheets and my finances or those things? So
- 22:11Speaker 2: I think everybody needs to know how to use AI
- 22:13Speaker 2: really efficiently. It's kind of like 20 years ago when
- 22:17Speaker 2: if you didn't know how to use Google really well,
- 22:19Speaker 2: because I couldn't remember everything in the world. So Google
- 22:23Speaker 2: came around to index it all. Now I can say
- 22:25Speaker 2: on Google easily, give me the answer to this question
- 22:29Speaker 2: or that question. The people who, I remember that being
- 22:32Speaker 2: almost a skill on resumes. I'm a really good Googler
- 22:37Speaker 2: where I can use Google. I can find answers really quickly.
- 22:40Speaker 2: And people would even respond to each other with emails
- 22:42Speaker 2: where it's like, hey, do you try Googling that before
- 22:44Speaker 2: answering the question? Because why are you bothering me with this?
- 22:47Speaker 2: Because you got enough from Google. Now people respond with, well,
- 22:49Speaker 2: did you chat GPT that? And so if you're not
- 22:54Speaker 2: able to do that, that's really it's the next version
- 22:56Speaker 2: of that Google skill really is the AI skill. It's harder.
- 23:00Speaker 2: It's really just systems engineering. And so I'm not sure
- 23:03Speaker 2: everyone's built or desires to be a systems engineer, but
- 23:06Speaker 2: that's really the skill which I think everybody needs to
- 23:09Speaker 2: do is become a systems engineer using AI with a
- 23:14Speaker 2: certain domain, whether it's accounting, whether it's software development, whether
- 23:17Speaker 2: it's HR, whatever it is. A systems engineer with AI
- 23:21Speaker 2: can get a lot done. And basically, you have to
- 23:25Speaker 2: strive to be the winner. It's brutal. I call it,
- 23:29Speaker 2: it's basically unbridled capitalism where each individual person has to,
- 23:34Speaker 2: or brutal capitalism, where each person has to worry about themselves.
- 23:38Speaker 2: And it's like, I have to be better than the
- 23:40Speaker 2: guy next to me because I know one of us
- 23:41Speaker 2: is going.
- 23:44Speaker 1: Yeah, I just saw that schools that were planning to
- 23:49Speaker 1: ban AI, Google said they would also ban Google searches
- 23:53Speaker 1: from the school too.
- 23:55Speaker 2: So, you know, then they'll be totally out of luck.
- 23:58Speaker 1: And speaking of that, how do you think we should
- 24:02Speaker 1: teach our children or the school systems to teach the
- 24:04Speaker 1: children if you put them in school now that AI
- 24:07Speaker 1: is here? Do we go completely AI, focus on their strengths?
- 24:11Speaker 1: How do we reinvent the learning system to prepare them
- 24:14Speaker 1: for the workforce?
- 24:16Speaker 2: Well, I think what's funny is we don't want to
- 24:19Speaker 2: reinvent it. All we have to do is go back
- 24:20Speaker 2: to the Greek times and say, well, what do they
- 24:22Speaker 2: do on the Greek square? And one of the things
- 24:25Speaker 2: I enjoy about my, my daughter went to university, Chicago,
- 24:27Speaker 2: for example, and Chicago, it's a classical liberal arts education,
- 24:32Speaker 2: especially for undergrad, where they teach you how to think,
- 24:34Speaker 2: they teach you how to argue, they teach you how to,
- 24:36Speaker 2: to process these things. That is what I think kids
- 24:40Speaker 2: should learn. And I feel that most teachers out there,
- 24:42Speaker 2: they teach you how to remember numbers like dates and
- 24:46Speaker 2: facts and, You know, get them right on the test.
- 24:49Speaker 2: So you go to the SAT test so you can
- 24:51Speaker 2: get them right on there. They don't teach you how
- 24:52Speaker 2: to think necessarily or how to argue and have that conversation.
- 24:57Speaker 2: Because knowledge has been obsoleted. It's a weird thing. Education
- 25:02Speaker 2: for a long time was viewed as the gaining of knowledge. Like, oh,
- 25:06Speaker 2: I know how to fix brakes on my car. Now
- 25:08Speaker 2: I'm a skilled mechanic. Well, I can fix brakes on
- 25:12Speaker 2: my car now by just asking ChatGPT to walk me
- 25:14Speaker 2: through it. Now, getting good at it with my hands,
- 25:16Speaker 2: getting dexterous enough to do it potentially might be a
- 25:18Speaker 2: skill I have to do there. But being able to
- 25:21Speaker 2: use that is the skill that's there. And so being
- 25:26Speaker 2: able to have that conversation, being able to think that through,
- 25:28Speaker 2: being able to have that logic and understand what to
- 25:31Speaker 2: do and how to do it and when to do it,
- 25:34Speaker 2: regardless of the knowledge, because the knowledge is free. It's
- 25:36Speaker 2: just there. Yeah. Teaching, you know, drilling facts into kids' heads,
- 25:40Speaker 2: which has been, when I was in high school, that
- 25:42Speaker 2: was a lot of it. It was just memorizing, memorization
- 25:45Speaker 2: of things, which made sense back then because the time
- 25:48Speaker 2: it took me to go to the library to look
- 25:49Speaker 2: it up in Encyclopedia Britannica was very long and very expensive.
- 25:53Speaker 2: So if I just had it at the tips of
- 25:54Speaker 2: my fingers, I had an advantage. It's the opposite now.
- 25:57Speaker 2: We have to just accept the fact that how we
- 26:01Speaker 2: learned and what we thought was valuable literally three years
- 26:04Speaker 2: ago or four years ago It's just not valuable anymore.
- 26:08Speaker 2: And I feel bad for people who are borrowing money
- 26:11Speaker 2: to go to college. And then four years later, they
- 26:13Speaker 2: get out of college and the thing which they just
- 26:17Speaker 2: paid $ 150, 000 for, $ 200, 000 for is already obsolete. So it's tough.
- 26:22Speaker 2: I think trade schools and pure logic schools or thinking
- 26:28Speaker 2: schools and the classical Greek philosophy type, schools are the
- 26:33Speaker 2: two forks that we need to have. We don't need
- 26:35Speaker 2: computer science schools. We don't need accounting schools for that reason.
- 26:39Speaker 2: You know, there's those factors there.
- 26:43Speaker 1: Yeah. It'll be interesting to see how each of the
- 26:47Speaker 1: school systems in the countries embrace going back to the
- 26:51Speaker 1: Greek times and different ways of learning and being a
- 26:54Speaker 1: critical thinker. How much is AI costing right now? And
- 26:58Speaker 1: OpenAI just halved some of its API prices. Meanwhile, XAI
- 27:02Speaker 1: and Grok increased prices because of their costs. Where do
- 27:07Speaker 1: you see the balance right now? And do you think
- 27:12Speaker 1: AI is cheap and it will increase in the future?
- 27:14Speaker 1: Or do you think it will get cheaper in the future?
- 27:18Speaker 2: Well, I would say AI is definitely very cheap right now.
- 27:22Speaker 2: If you look at free AI, for example, if you
- 27:24Speaker 2: go to Google's website and you search for something, you
- 27:26Speaker 2: can use their agent mode. That's free with some ads
- 27:29Speaker 2: and you can ask questions there. even like my $ 200
- 27:33Speaker 2: a month subscription to cloud or to Codex, what I
- 27:36Speaker 2: can get done with that compared to a human, it's
- 27:40Speaker 2: basically a $ 200 a month subscription is equivalent to a $ 200, 000
- 27:45Speaker 2: a year engineer, or maybe two of them or three
- 27:49Speaker 2: of them. I don't know. But that exchange, if you
- 27:51Speaker 2: think about it, is mind-bogglingly good as far as it's cheap. However,
- 27:56Speaker 2: having said that, There's a competition between all these different companies.
- 28:00Speaker 2: So if I can get that same quality of work
- 28:02Speaker 2: for $ 100 from somebody else, well, I'm going to have
- 28:05Speaker 2: to pay $ 200 if I don't have to, right? So
- 28:07Speaker 2: there's the cost of goods. If you think about TVs,
- 28:11Speaker 2: a 65-inch TV costs like $ 500 at Walmart right now.
- 28:15Speaker 2: That's really cheap, but at the same time, it costs
- 28:17Speaker 2: not that much to make it, so why not? And
- 28:19Speaker 2: so it's... It's strange as far as the competition and
- 28:25Speaker 2: the capitalism of AI has driven the price down. But
- 28:28Speaker 2: the ROI of paying for AI, even if you had
- 28:32Speaker 2: to pay the token prices instead of subscription prices, is
- 28:34Speaker 2: still very, very good. Assuming you know how to use
- 28:36Speaker 2: it and you actually get the 5x, 10x benefit out
- 28:39Speaker 2: of it. So I would say it's actually very cheap.
- 28:41Speaker 2: And I think that's one of the mistakes that these
- 28:43Speaker 2: big companies made was they did that calculation. They said, well,
- 28:47Speaker 2: clearly a person would pay... $ 1, 000 a month to replace
- 28:51Speaker 2: a $ 5, 000 a month employee, right? It's 20% of the cost.
- 28:57Speaker 2: But if the next guy comes around and says, well,
- 28:58Speaker 2: I can do it for $ 500 or I can do
- 29:00Speaker 2: it for $ 300 or I can do it for $ 200
- 29:01Speaker 2: or for $ 100, everyone's going to go for the cheaper one.
- 29:04Speaker 2: And the cost of goods for that is going to
- 29:06Speaker 2: go down really, really fast as technology gets better. We've
- 29:09Speaker 2: already seen this with coding where the frontier models are
- 29:13Speaker 2: really good at coding, but the Two or three generation
- 29:16Speaker 2: old models are also really very good at coding, especially
- 29:19Speaker 2: if you're doing something like making a website. It's not
- 29:21Speaker 2: that hard. It's a known thing. So I don't need
- 29:23Speaker 2: to use the latest and greatest fable model out there
- 29:28Speaker 2: to make a website. It's very doable with Muse or
- 29:33Speaker 2: some cheaper website builder out there. So it's cheap from
- 29:38Speaker 2: the standpoint of ROI. but it's expensive from the standpoint
- 29:42Speaker 2: of it's going to get much cheaper faster.
- 29:48Speaker 1: Yeah, I was mind blown last month. I bought a
- 29:51Speaker 1: domain and built a website and it was public and live,
- 29:54Speaker 1: fully done within five minutes. A lot of people are
- 29:57Speaker 1: starting to rely on the AI. And although we have
- 30:01Speaker 1: the competitiveness that's bringing the price down, what if we're
- 30:07Speaker 1: relying on it and then they increase the price? Do
- 30:10Speaker 1: we switch to open source models? Do we just bite it?
- 30:12Speaker 1: What do you think will happen if that happens?
- 30:16Speaker 2: Well, I think that's already starting to happen for a
- 30:17Speaker 2: couple of reasons. One is because of privacy. Obviously, people
- 30:20Speaker 2: are concerned about their code, for example, or their business
- 30:23Speaker 2: ideas or their product ideas going to the cloud when
- 30:26Speaker 2: they're using these frontier models. So people are going out
- 30:29Speaker 2: and buying a $ 5, 000 or $ 10, 000 box from NVIDIA. They
- 30:33Speaker 2: can put it on their desktop and it's got 128
- 30:36Speaker 2: or 256 gigabytes of RAM. It can run a really
- 30:39Speaker 2: good open source model that can code for you and
- 30:41Speaker 2: make that website in maybe 10 minutes instead of five minutes.
- 30:44Speaker 2: But it sits on your desk and it's a fixed cost.
- 30:46Speaker 2: Once you pay your $ 5, 000 or $ 10, 000, you got to
- 30:48Speaker 2: pay your $ 5 or $ 10 a month in electricity that
- 30:51Speaker 2: runs it. And it's sovereign. So I think there's going
- 30:55Speaker 2: to be a shift to sovereign AI. I'm a big
- 30:59Speaker 2: pro-sovereign guy. I'm sure you know that. So sovereignty, I think,
- 31:02Speaker 2: is going to be an interesting part of this. And
- 31:04Speaker 2: I think it's going to be hard for these big
- 31:06Speaker 2: companies out there to be able to try and monopolize
- 31:08Speaker 2: or collude the price up because the catch-up of the
- 31:13Speaker 2: open models is so fast that for them to provide
- 31:17Speaker 2: benefits from their models that are so much superior to
- 31:21Speaker 2: the open source models that you can run locally is
- 31:24Speaker 2: just not there for most people. And so I think
- 31:28Speaker 2: what's the mode and what's the value add is going
- 31:30Speaker 2: to be a question which all these companies have to
- 31:32Speaker 2: ask themselves on an ongoing basis because they're showing the
- 31:36Speaker 2: modes are very hard to defend because people just copy
- 31:39Speaker 2: or learn from your models. The value add right now
- 31:43Speaker 2: is tooling in many cases, but even that, if you
- 31:45Speaker 2: look at OpenCode and some of those other IDEs and
- 31:50Speaker 2: platforms and the tools on your computer, they run quite well.
- 31:55Speaker 2: I would say it Codex and Grox and Cloud's tool
- 32:00Speaker 2: set that they use for the AI is still superior
- 32:03Speaker 2: and it makes it run better for you and can
- 32:05Speaker 2: do more things and use less tokens. But there's definitely,
- 32:08Speaker 2: you know, in an apocalyptic model where, you know, AI's
- 32:13Speaker 2: data centers go away and you have your NVIDIA device
- 32:15Speaker 2: sitting at home with your laptop, you can still code
- 32:17Speaker 2: just fine.
- 32:20Speaker 1: Do you think companies should be concerned for the frontier models?
- 32:25Speaker 1: with their proprietary information going in there and looking at
- 32:29Speaker 1: sovereign AI options? And does XYO pursue that?
- 32:34Speaker 2: So, yes, I think there should be concern, especially when
- 32:37Speaker 2: it comes to private information, for example, whether it's social
- 32:43Speaker 2: security numbers or those sorts of things. I think IP
- 32:48Speaker 2: people worry about where there's a patent or there's those
- 32:51Speaker 2: sorts of things. One of the things that's really hard
- 32:53Speaker 2: to do is if... AI learns from the feed that
- 32:58Speaker 2: you've basically given it. It just knows things. It's kind
- 33:02Speaker 2: of like an apprentice learning from a master, and the
- 33:04Speaker 2: apprentice just knows things. Well, is that copying it? Well,
- 33:08Speaker 2: not really. It just happens to know how to make
- 33:10Speaker 2: a sword really well, right? So if this thing learned,
- 33:14Speaker 2: for example, how to make chips better because it watched
- 33:18Speaker 2: your feed and it trained on that, And now you
- 33:20Speaker 2: ask it, well, make me, design me a new ARM chip.
- 33:24Speaker 2: And it just happens to know how to do that.
- 33:25Speaker 2: Is that copying your IP or not? It's a lot
- 33:30Speaker 2: harder than it used to be because really what you're
- 33:32Speaker 2: doing now is defending against effectively the employee that defects
- 33:36Speaker 2: that happens to have honed their skill at your company
- 33:39Speaker 2: and now can take that skill somewhere else. You've seen
- 33:40Speaker 2: that with SpaceX, for example. There's people who are at
- 33:42Speaker 2: SpaceX and they left and they go and start their
- 33:44Speaker 2: own space company or a different space company. We've always
- 33:47Speaker 2: decided that that's okay because once a person has that
- 33:49Speaker 2: skill and they've learned that skill, that's theirs to go
- 33:52Speaker 2: and monetize and to use and to promote. That's kind
- 33:56Speaker 2: of what they're providing to those AIs. They're helping the
- 33:59Speaker 2: AI get better and better and better. And then the
- 34:01Speaker 2: AI basically available to everyone else out there to go
- 34:03Speaker 2: compete with them. But it's hard to keep that genie
- 34:09Speaker 2: in the bottle, right? So even if you have employees
- 34:12Speaker 2: who are just asking questions on their own personal AI,
- 34:15Speaker 2: Those AIs will learn about that or those AIs read
- 34:18Speaker 2: all the white papers out there and all the documents
- 34:19Speaker 2: out there and all the research out there. So preventing
- 34:23Speaker 2: AI from basically learning things that are going on in
- 34:26Speaker 2: the world, keeping that private in your company, keeping IP private,
- 34:31Speaker 2: keeping product plans private is so much harder than it
- 34:35Speaker 2: used to be already. I'm not sure that juice is
- 34:39Speaker 2: worth the squeeze to not have the benefit of AI
- 34:43Speaker 2: just for the fear of having your private information getting
- 34:47Speaker 2: out there.
- 34:48Speaker 1: Yeah. I recently did an interview with an AI and
- 34:54Speaker 1: robotics and blockchain company. And one of the things I
- 34:58Speaker 1: took away from that was that software has been solved
- 35:00Speaker 1: and hardware is where the smart people are working right now.
- 35:04Speaker 1: And as machines and IoT devices get more prevalent, I'm
- 35:08Speaker 1: curious how that fits into XYO and.
- 35:12Speaker 1: How much of a focus is on hardware and tying
- 35:15Speaker 1: the software into it.
- 35:18Speaker 2: I think you're very right there as far as software
- 35:20Speaker 2: has been solved. I wouldn't say it's been completely solved,
- 35:22Speaker 2: but it's been solved for the vast majority of the
- 35:26Speaker 2: use cases. I don't think hardware is that far behind.
- 35:29Speaker 2: I think hardware is getting close to being solved also, right?
- 35:32Speaker 2: It's for me to go and design a new chip
- 35:34Speaker 2: or a new PCB, it takes now minutes as opposed to,
- 35:39Speaker 2: I've used to be in the hardware business and it
- 35:42Speaker 2: would take months for a hardware engineer to go lay
- 35:44Speaker 2: out the PCB. And then, you know, it takes some
- 35:47Speaker 2: iterations because after they do it, it's like, oh, it
- 35:49Speaker 2: turns out that, you know, these two traces harmonize and
- 35:52Speaker 2: cause interference on each other and so on and so forth.
- 35:54Speaker 2: But the AI can detect those things way faster. So
- 35:56Speaker 2: it's like circuit board design, chip design, those sorts of
- 36:00Speaker 2: things have already been done with AI quite a bit,
- 36:05Speaker 2: very successfully. I would even argue to some degree those
- 36:09Speaker 2: things fall, they're kind of hard software, if that makes sense,
- 36:12Speaker 2: because it's really a logic board. This just happens to
- 36:15Speaker 2: be burnt into something. It's not really hardware. The hardware
- 36:19Speaker 2: that's probably harder are like the robotic arms and the
- 36:21Speaker 2: hands and the physical attributes and being able to move things.
- 36:28Speaker 2: That also is something which I'm sure they're using a
- 36:30Speaker 2: lot of AI to simulate those. We've got great 3D
- 36:35Speaker 2: physics engines, for example, you can go and test these
- 36:38Speaker 2: things prior to building them. So you can go and
- 36:40Speaker 2: design things, test them in a 3D physics engine on
- 36:42Speaker 2: a computer and software, and get 99% of the problems
- 36:46Speaker 2: out of it. Then go build a prototype. The hard
- 36:48Speaker 2: parts of the materials, maybe the materials didn't quite get
- 36:50Speaker 2: modeled correctly in the simulation, those are going to get
- 36:54Speaker 2: solved as well really fast there. And I think there's
- 36:57Speaker 2: a lot of opportunity there, especially in the short term,
- 37:00Speaker 2: for people to innovate and make some money there. As
- 37:03Speaker 2: you said, the smart people are going there. I think
- 37:06Speaker 2: it's going to be a lucrative situation for them. I
- 37:12Speaker 2: look at AI in general as an opportunity. I think
- 37:14Speaker 2: a lot of people look at it as this fear
- 37:15Speaker 2: because of the fact it's going to take away what
- 37:17Speaker 2: they had. I look at it as an opportunity because
- 37:20Speaker 2: every morning you wake up and you're like, well, I
- 37:21Speaker 2: can do something today that I didn't think I could do.
- 37:24Speaker 2: I couldn't possibly have done before. Only thing is you
- 37:27Speaker 2: have to do it fast because if you wait six
- 37:29Speaker 2: weeks or eight weeks, somebody else will do it first.
- 37:31Speaker 2: And so, Jumping on things, the whole world's just moving really, really,
- 37:35Speaker 2: really fast, which is scary for people. People are used
- 37:39Speaker 2: to having, I got a skill in college, I worked
- 37:42Speaker 2: for 40 years, got a pension, I retired, life's good.
- 37:46Speaker 2: That life's gone. I think people have to just realize,
- 37:48Speaker 2: you can argue whether that's a good thing or a
- 37:50Speaker 2: bad thing. I don't know if it's a good thing
- 37:52Speaker 2: or a bad thing, but that life is gone. It
- 37:54Speaker 2: just is.
- 37:56Speaker 1: Yeah, I agree. And what do you think... is the
- 38:01Speaker 1: main thing or things that need to be solved before
- 38:04Speaker 1: we start seeing millions of humanoid robots acting and doing
- 38:09Speaker 1: service for us in the real world? Because we've seen
- 38:11Speaker 1: some already, but it's very early stage. They're very clunky.
- 38:14Speaker 1: What are those main things we need to solve?
- 38:17Speaker 2: For me personally, I think one of the biggest questions
- 38:19Speaker 2: is a sovereignty question. So, well, who really controls it?
- 38:22Speaker 2: And we've already seen this to some degree with Tesla
- 38:24Speaker 2: cars where, well, Tesla gets stolen and then they can
- 38:27Speaker 2: call up Tesla and Tesla can disable the car remotely. Well,
- 38:32Speaker 2: that means that an AI that's running on Tesla's servers
- 38:35Speaker 2: can also disable cars remotely, right? I remember a speaker
- 38:40Speaker 2: at a crypto conference years ago who made this comment
- 38:45Speaker 2: about the difference between can't and won't and how it's important.
- 38:48Speaker 2: I think his comment was, what's the difference between can't
- 38:50Speaker 2: and won't? And he said, 20 to life, I think
- 38:53Speaker 2: was the answer. Because if you go to the government
- 38:55Speaker 2: and you say, well, do this, and you're like, well,
- 38:57Speaker 2: I won't, then they can throw you in jail. If
- 38:59Speaker 2: you say you can't, well, what are they going to do?
- 39:01Speaker 2: it's meant to be can't. And that's one of the
- 39:03Speaker 2: things that scared me a little bit about this Binance
- 39:05Speaker 2: hack was the fact that they could freeze about half
- 39:08Speaker 2: those stablecoins, the fact that they could freeze those. It's
- 39:11Speaker 2: good that they froze them from the standpoint that, well,
- 39:12Speaker 2: the perpetrator didn't get away with them, but it's bad
- 39:14Speaker 2: that they could freeze them because, well, who has the
- 39:17Speaker 2: right to freeze those? Which is kind of a strange question.
- 39:20Speaker 2: And so I think they're going to have to solve
- 39:22Speaker 2: that answer of, well, who has the right to be
- 39:26Speaker 2: able to control those robots? Is that something which has
- 39:29Speaker 2: a centralised control system? Is that something which Tesla can
- 39:32Speaker 2: go and pull the license for something? It's like, oh,
- 39:35Speaker 2: I saw that your robot was digging in your backyard,
- 39:38Speaker 2: doing something it's not supposed to do. I'm going to
- 39:40Speaker 2: turn it off because that violates our terms. I think
- 39:45Speaker 2: we have to answer that question of if I buy
- 39:47Speaker 2: a robot, Do I own it or is this something
- 39:50Speaker 2: which I'm leasing from a company and they have the
- 39:53Speaker 2: right to tell me what I can or cannot do
- 39:54Speaker 2: with it? We see the same thing in Ukraine right
- 39:56Speaker 2: now where the policies of what can you use Starlink
- 40:02Speaker 2: for are what they are. And if they violate the policies,
- 40:05Speaker 2: they can get turned off. Is that something which people want?
- 40:10Speaker 2: I don't think so. I would rather have a robot
- 40:12Speaker 2: in my house that I know I control and I
- 40:15Speaker 2: have final say over as opposed to a robot that
- 40:18Speaker 2: is in the cloud. And I think that's going to
- 40:20Speaker 2: also drive some cost things because I'm going to need
- 40:22Speaker 2: to onboard GPU for it. So it runs and it
- 40:25Speaker 2: doesn't do things. You still have the problem of, well,
- 40:27Speaker 2: it needs to get updates. It needs to go and
- 40:29Speaker 2: get patched. And so, well, where do I get the
- 40:31Speaker 2: update from? So it's, it's a really complicated question, but
- 40:34Speaker 2: I think that the sovereignty question of who controls it,
- 40:38Speaker 2: does government has, or do governments have back doors to
- 40:40Speaker 2: it to be able to turn them on or off
- 40:42Speaker 2: or to monitor them? Um, Who controls the AI? Who
- 40:46Speaker 2: owns the AI? Specifically when it comes to physical AI
- 40:49Speaker 2: as opposed to software AI is going to be a
- 40:53Speaker 2: huge question. And I don't think people appreciate the complexity
- 40:57Speaker 2: of that. It's not as simple as let's just pass
- 40:58Speaker 2: a law that says blank. It's very, very nuanced.
- 41:05Speaker 1: Yeah, it's so true. And with everything moving to subscriptions
- 41:11Speaker 1: and sort of like you're always tied to the company
- 41:13Speaker 1: that produced it. it's becoming obsolete where you buy something
- 41:17Speaker 1: and you own it, you know?
- 41:19Speaker 2: Right. Well, video games are that way. Like you don't,
- 41:21Speaker 2: you buy a license to a video game that, you know,
- 41:23Speaker 2: used to be, I have a physical cartridge and like
- 41:24Speaker 2: the cartridge you have forever. It's like, I guess they'll
- 41:26Speaker 2: play my Atari 2600 games today. Um, they work, they're cartridges.
- 41:30Speaker 2: They have little things on there on the chip. I
- 41:32Speaker 2: can't play my, you know, some of my, um, Xbox
- 41:36Speaker 2: games or my Xbox 360 games because they've been sunset
- 41:39Speaker 2: by their store. I can't watch the movies on Prime
- 41:43Speaker 2: anymore because they've lost the license to that movie I
- 41:46Speaker 2: quote-unquote bought or I own on their thing. But I
- 41:49Speaker 2: don't really own it because once they lose the license,
- 41:51Speaker 2: they're not allowed to stream it to me anymore. So
- 41:54Speaker 2: physical Blu-rays, even in those cases, they have the encryption.
- 41:58Speaker 2: So if I don't have the right to decrypt it
- 42:03Speaker 2: or whatever it is, I don't really own it. So
- 42:05Speaker 2: the definition of ownership, the definition of sovereignty, in a
- 42:08Speaker 2: strange way, going back to the the Greek logic and
- 42:12Speaker 2: even going back to the early American days where you
- 42:15Speaker 2: have your castle, the man's castle, where it's like, here's
- 42:20Speaker 2: my property, here's my house. What goes on inside there
- 42:22Speaker 2: is my business sort of a thing. I think people
- 42:24Speaker 2: are going to revert back to that mentality of, I
- 42:28Speaker 2: want to have control over my life and my things
- 42:33Speaker 2: because of the fact that they're going to feel so
- 42:35Speaker 2: out of control.
- 42:38Speaker 1: Amazing, Ari.
- 42:39Speaker 2: One last question.
- 42:40Speaker 1: Regarding the future as AI develops, both the software side
- 42:45Speaker 1: and more so the hardware side. What's your vision for
- 42:49Speaker 1: XYO and integrating and bringing that sovereignty to ensure that
- 42:54Speaker 1: hopefully I can own my own robot and all of
- 42:56Speaker 1: the other machines that will be tied into these networks?
- 43:02Speaker 2: Well, the thing that's the most exciting for me is
- 43:04Speaker 2: the fact that I have AI to be able to
- 43:05Speaker 2: help me build these things a lot faster, right? So I...
- 43:09Speaker 2: I've never had a shortage of ideas. I think a
- 43:11Speaker 2: lot of entrepreneurs have that. People are like, oh, do
- 43:12Speaker 2: you have any good ideas? I have a hundred good ideas.
- 43:14Speaker 2: I just don't have the ability to make them all.
- 43:16Speaker 2: And Elon Musk actually is very fortunate because he can
- 43:19Speaker 2: make his boring company, he can make his space, he
- 43:21Speaker 2: can make all these different things because he has so
- 43:22Speaker 2: much money and he dedicates so much time to be
- 43:25Speaker 2: able to try all these different things. With AI now,
- 43:28Speaker 2: I feel that the ability for us to go and
- 43:30Speaker 2: try and fail with more things is something which is great,
- 43:33Speaker 2: which is very exciting for us. But I think the
- 43:35Speaker 2: focus of XBO and XL1 will remain... the data-centric view
- 43:41Speaker 2: of reality. And I think I described, it's a project
- 43:44Speaker 2: called Webble, which I've been working on for years, which
- 43:46Speaker 2: will come out in the not so distant future. But
- 43:49Speaker 2: the idea of a Webble is that each person's reality,
- 43:51Speaker 2: or even for that reason, each item's reality is just like,
- 43:54Speaker 2: it's identity with all of its history combined. That's your Webble, right?
- 43:59Speaker 2: So the definition of being Ashton is basically it's your memories.
- 44:04Speaker 2: It's where you are right now and your memories. 10
- 44:07Speaker 2: years ago, you were a different Ashton. You're still Ashton,
- 44:09Speaker 2: but you were a different Ashton. Now you're, you know, you've,
- 44:11Speaker 2: you've grown certain ways, you've lost certain things or whatever
- 44:13Speaker 2: it is. Now you're a different Ashton, but that's, that's
- 44:16Speaker 2: what defines you is that, that bubble and that kind
- 44:18Speaker 2: of concept that's there. And that's the way I look at,
- 44:22Speaker 2: at the whole world that each item, each thing is
- 44:25Speaker 2: sovereign and they, you know, they are a reality onto themselves.
- 44:29Speaker 2: And all we can do really is observe those. I
- 44:32Speaker 2: can't go and see what you're thinking, but I can,
- 44:36Speaker 2: infer what you're thinking by observing you with a camera
- 44:38Speaker 2: or whatever it is. And same kind of thing with robots.
- 44:40Speaker 2: And so the communications and the interactions between these levels
- 44:45Speaker 2: and these realities of all these different things is something
- 44:48Speaker 2: which really fascinates me. And I think with AI, the
- 44:52Speaker 2: ability to actually capture that and to analyze that and
- 44:56Speaker 2: to see that and to use that for security, use
- 44:59Speaker 2: that for advancement is something which is now available to us.
- 45:03Speaker 2: So I feel like something I've worked on for 10
- 45:04Speaker 2: years is finally possible, if that makes sense. Amazing.
- 45:10Speaker 1: I'd love to see your ideas come to fruition. I'll
- 45:14Speaker 1: definitely be checking out that personal project as it comes out.
- 45:18Speaker 1: I appreciate your insights into everything going on with AI
- 45:21Speaker 1: right now. I think this is the deepest we've been
- 45:23Speaker 1: on all our talks so far, and there's still a
- 45:26Speaker 1: thousand questions I have. I'll save for our next episode together.
- 45:30Speaker 1: Wishing you and the team all the best, Ari, moving forward,
- 45:33Speaker 1: and would love to follow up again in the very
- 45:35Speaker 1: near future.
- 45:36Speaker 2: Thank you very much, Ashton, for having me on.