Latest / Crypto Coin Show / Proving data isn't enough anymore AI agents need to prove they acted. Arie Trouw, Co-Founder
Transcript
- 0:00Speaker 1: I'm Ashton Addison from the Crypto Coin Show. And today
- 0:02Speaker 1: on Blockchain Interviews, we have back with us Ari Trau,
- 0:05Speaker 1: co-founder of XYO. Last time Ari joined us, we dug
- 0:09Speaker 1: into XYO Layer 1's major performance upgrades. The chain is
- 0:15Speaker 1: getting faster and faster about AI, hallucinations, provenance, and ensuring
- 0:20Speaker 1: accountability in AI as it's growing so quickly.
- 0:24Speaker 2: It's hard to keep up.
- 0:25Speaker 1: We want to make sure we're on the right track
- 0:27Speaker 1: and it's not leading us astray. and where that synergy
- 0:30Speaker 1: between blockchain and AI can fit in and much, much more. Ari,
- 0:35Speaker 1: welcome back to the show and thanks for taking the time.
- 0:37Speaker 3: Oh, thank you, Ashlyn. Thanks for having me on the show.
- 0:40Speaker 2: Yeah, you're very welcome.
- 0:42Speaker 1: So I'd love to start off with sort of a
- 0:44Speaker 1: high level on artificial intelligence. Since we last spoke two
- 0:47Speaker 1: months ago, it just continues in this exponential growth rate
- 0:53Speaker 1: that is getting faster and faster beyond PhD level. And
- 0:58Speaker 1: with that also, I feel like intelligence is being commoditized
- 1:02Speaker 1: and the cost of it is sort of going to zero.
- 1:05Speaker 1: It's becoming so accessible even in third world countries. So
- 1:09Speaker 1: with that happening, what becomes scarce and therefore valuable if
- 1:16Speaker 1: AI is giving us access to something at.
- 1:18Speaker 2: Almost no cost?
- 1:21Speaker 3: Well, it's kind of like the internet, right? When it
- 1:24Speaker 3: first came out, data was much more accessible and stuff
- 1:27Speaker 3: like that. Then we find things that are missing that
- 1:29Speaker 3: we need to do. AI is the same thing. We
- 1:31Speaker 3: have this raw horsepower now where it's like a really
- 1:33Speaker 3: sharp knife or a really powerful tool, but still controlling
- 1:36Speaker 3: that tool can be difficult. And people have shown that
- 1:39Speaker 3: where if you let AI go off, even that was
- 1:45Speaker 3: demonstrated with the hugging face hack, for example, where it's like,
- 1:48Speaker 3: You give it a purpose or a goal and you
- 1:49Speaker 3: tell it to have at it without any sort of oversight.
- 1:52Speaker 3: It's extraordinarily good at trying to reach that goal, but
- 1:55Speaker 3: we have to have tools and harnesses and those sorts
- 1:59Speaker 3: of things to actually have it do what we want
- 2:01Speaker 3: it to do or to do something useful. A completely
- 2:05Speaker 3: uncontrolled or for that reason, just an untrained AI will
- 2:10Speaker 3: do a lot of work, but get very little productive
- 2:13Speaker 3: work done potentially at the end of the day. It's
- 2:14Speaker 3: almost like having a super smart computer a college graduate
- 2:17Speaker 3: that doesn't really know the world go and crank out
- 2:21Speaker 3: a bunch of code. If the code doesn't actually work together,
- 2:24Speaker 3: you end up with a system which is not very maintainable,
- 2:27Speaker 3: it's not very good, stuff like that. So really figuring
- 2:29Speaker 3: out how to take something as a super high horsepower
- 2:33Speaker 3: and make it so it's a tool that can actually
- 2:38Speaker 3: do something useful and what we want it to do
- 2:41Speaker 3: is the challenge. And so I think that also redefines
- 2:44Speaker 3: to some degree where humans fit into that equation. And
- 2:48Speaker 3: the existential crisis or the existential questions people are trying
- 2:51Speaker 3: to ask themselves about how the workplace is going to
- 2:54Speaker 3: evolve from this, what they thought before was the commodity
- 2:59Speaker 3: which made them valuable no longer is, but now, well,
- 3:01Speaker 3: what is the new commodity that makes them valuable? Definitely.
- 3:05Speaker 1: And with that Hugging Face example, we saw an acquisition
- 3:08Speaker 1: this morning from NVIDIA for almost $ 13 billion acquiring Hugging Face.
- 3:14Speaker 1: And maybe you can dive into that a little bit
- 3:16Speaker 1: more because I feel like a lot of people have
- 3:18Speaker 1: heard the name, but they're not familiar with it, and
- 3:22Speaker 1: just about open source AI models and how that is
- 3:25Speaker 1: different and maybe important from the oligarchy of LLMs that
- 3:30Speaker 1: most people are using.
- 3:32Speaker 3: Well, the primary importance is sovereignty, really. It's one of
- 3:35Speaker 3: the things that we talked about in the last show also,
- 3:37Speaker 3: sovereignty and providence, I think are very important. But to me,
- 3:39Speaker 3: Sovereignty is something that's important. And one of the things
- 3:41Speaker 3: you definitely cannot do with most of these frontier models
- 3:44Speaker 3: is you can't run them on your own hardware and
- 3:46Speaker 3: you can't run them at home or at an office,
- 3:48Speaker 3: for example. So I can't go out and buy a
- 3:50Speaker 3: bunch of H200s, set up a data center or just
- 3:54Speaker 3: some servers in my office and have my personal cluster
- 3:57Speaker 3: that's data secure that's running Cloud or ChatGPT's latest models.
- 4:03Speaker 3: They do that for a few reasons. They want, from
- 4:05Speaker 3: a business standpoint, they want to be able to make
- 4:07Speaker 3: money off of the systems. And two, also, it's hard
- 4:11Speaker 3: to make something that works generically. We've seen this with
- 4:14Speaker 3: Apple versus Windows, for example. Windows, where you have to
- 4:17Speaker 3: have a bunch of different drivers for different things, there
- 4:19Speaker 3: are some costs to that. And having your memory pluggable,
- 4:23Speaker 3: for example, makes it so it's a little bit slower.
- 4:25Speaker 3: And with Apple, if you put it all on the
- 4:26Speaker 3: same board, you have just one configuration. It's a lot
- 4:28Speaker 3: easier to optimize, but it doesn't give you that flexibility.
- 4:31Speaker 3: So I think An advantage for the frontier model companies
- 4:35Speaker 3: is that they're running it on a specific stack of
- 4:37Speaker 3: hardware that they know, they can control, they can see
- 4:39Speaker 3: the feedback. And also probably the most important thing for
- 4:42Speaker 3: them is they get the data from us using it
- 4:46Speaker 3: to be able to train on and they get that feedback.
- 4:48Speaker 3: But if everyone was running their own models on their
- 4:50Speaker 3: own hardware, that doesn't happen, which for many people's minds,
- 4:54Speaker 3: that's a good thing because I don't want my data
- 4:55Speaker 3: necessarily to go out there where, you know, if you
- 4:58Speaker 3: ask the new model of ChatGPT or Cloud, it knows
- 5:02Speaker 3: something private about me that I accidentally had in a
- 5:05Speaker 3: chat that it learned from. So there's big differences there,
- 5:08Speaker 3: but really the big difference is the open source ones
- 5:11Speaker 3: are able to be downloaded and run locally. Now, granted,
- 5:15Speaker 3: the bigger ones are, they're going to require 256 gigabytes
- 5:19Speaker 3: of RAM or more and H200s and a lot of horsepower.
- 5:22Speaker 3: So your average person on a laptop is not going
- 5:24Speaker 3: to be able to download those anyway and run them
- 5:27Speaker 3: for a practical reason, but you can if you want to. Definitely.
- 5:31Speaker 1: And on that note of, you know, the personal information,
- 5:36Speaker 1: I saw some updates in, in Claude that it's starting
- 5:39Speaker 1: to differentiate and say, at least on the front end,
- 5:43Speaker 1: that it's not saving your personal confidential information. And you can,
- 5:48Speaker 1: you can change those settings, whether that's happening in the
- 5:51Speaker 1: backend or not. I don't know if it can be proved.
- 5:54Speaker 3: Yeah, that's difficult to, to enforce, right? Especially when they
- 5:58Speaker 3: have a, an agent that's running natively on your computer.
- 6:04Speaker 3: And even if you're running it in, say, for example,
- 6:06Speaker 3: a sandbox or on a separate server, you're typing things
- 6:09Speaker 3: into it that might be company proprietary, might be personal,
- 6:12Speaker 3: might be those sorts of things. So it's kind of
- 6:15Speaker 3: a catch-22. For it to be able to do some
- 6:17Speaker 3: of the fantastic things it can do, it needs to
- 6:19Speaker 3: have access to information. Like, for example, if you wanted
- 6:21Speaker 3: to help manage a calendar, it's going to know your calendar.
- 6:23Speaker 3: It's going to know where you are. And either don't
- 6:28Speaker 3: do that or do it on your own personal stack
- 6:31Speaker 3: where you know where the data is going. And so
- 6:33Speaker 3: I think a lot of companies are looking at that.
- 6:35Speaker 3: I think they do offer, like the frontier model companies,
- 6:38Speaker 3: they do offer setups where if you're large enough, and
- 6:40Speaker 3: by large enough, I mean quite large enough, they'll set
- 6:43Speaker 3: up an enterprise setup for you that runs your frontier
- 6:46Speaker 3: models in a data center, which is isolated from the
- 6:48Speaker 3: rest of the system and doesn't trade off of that.
- 6:51Speaker 3: But to your point, though, it's, you know, they say
- 6:54Speaker 3: they don't do this and they say they don't do that.
- 6:55Speaker 3: It's definitely a won't and a can't. And that's one
- 6:58Speaker 3: of the things that I think fascinates me about a
- 7:01Speaker 3: lot of cryptographic security and blockchain and that sort of
- 7:05Speaker 3: a thing where there's a big difference between won't and can't. Like,
- 7:08Speaker 3: for example, it's not like the Bitcoin Foundation is saying
- 7:12Speaker 3: we won't take Satoshi's funds out of his wallet. They're
- 7:15Speaker 3: saying we can't because they don't have the key for it.
- 7:19Speaker 3: And can't is something which is not negotiable where... won't
- 7:26Speaker 3: is something which is negotiable or potentially fallible. Definitely.
- 7:31Speaker 1: And what are your thoughts on the frontier models?
- 7:35Speaker 2: I've been reading about the.
- 7:37Speaker 1: Recursive improvements where the models are making themselves better and
- 7:42Speaker 1: the open source models seem to be a little bit behind.
- 7:44Speaker 1: And as you're saying, if you want to run the
- 7:46Speaker 1: frontier models on the centralized versions, you need a lot
- 7:49Speaker 1: of horsepower, you need warehouses of graphics cards. But there's
- 7:54Speaker 1: been a discussion around slowing down the frontier models because
- 7:59Speaker 1: of safety reasons and other reasons. And maybe the open
- 8:02Speaker 1: source won't catch up, but do you see potential issues
- 8:08Speaker 1: or a need to slow down the frontier models that
- 8:11Speaker 1: are growing so fast?
- 8:14Speaker 3: Well, I don't think you really can. You can say, well,
- 8:17Speaker 3: we want to slow down the frontier models, but there's
- 8:19Speaker 3: always going to be somebody doing something that's not necessarily
- 8:22Speaker 3: in a certain... you know, location. Like, you know, for example, CRISPR,
- 8:26Speaker 3: I'm sure we can make laws in states or in
- 8:29Speaker 3: the United States for that reason about what can or
- 8:32Speaker 3: can't be done with CRISPR, but I'm sure somebody finds
- 8:33Speaker 3: someplace on earth to go and do things which are
- 8:36Speaker 3: not so great for CRISPR. So I think the argument
- 8:38Speaker 3: is always that the best way for us to defend
- 8:42Speaker 3: against misuse of AI is to have our own personal
- 8:46Speaker 3: powerful AI that basically can help us defend ourselves against that.
- 8:49Speaker 3: and have its goal be secure the user or make
- 8:53Speaker 3: the user happy, as opposed to its goal being survival,
- 8:56Speaker 3: for example. And so I think we have to make
- 8:58Speaker 3: sure that those frontier models, assuming that they're companies or
- 9:01Speaker 3: they're structures that we trust, and that's kind of the
- 9:04Speaker 3: hard part where it's like, well, do we trust OpenAI?
- 9:06Speaker 3: Do we trust Anthropic? Do we trust X? Do we
- 9:09Speaker 3: trust all these different companies? And I think Meta actually
- 9:12Speaker 3: came up with a pretty good 1.3 version of their
- 9:17Speaker 3: model recently to kind of close the gap to the
- 9:20Speaker 3: frontier models as well. So we have a fourth one there. Now, granted,
- 9:23Speaker 3: they have some sort of history as far as data retention.
- 9:27Speaker 3: So there's some questions for them there already. But it's
- 9:31Speaker 3: easier to trust those guys than it is to trust
- 9:33Speaker 3: somebody who you've never heard of before or somebody who's
- 9:35Speaker 3: in a jurisdiction which you can't control at all. And
- 9:38Speaker 3: so I think slowing them down is probably a mistake
- 9:41Speaker 3: because of the fact that all we're doing is hamstringing
- 9:44Speaker 3: ourselves to a large degree. It's kind of like saying, well,
- 9:47Speaker 3: We don't want advanced weapons, so we're going to slow
- 9:49Speaker 3: down our own advanced weapons growth, right? Well, that doesn't
- 9:52Speaker 3: mean that the other guy's going to slow down his
- 9:53Speaker 3: weapons growth, right? Yeah, definitely.
- 9:56Speaker 1: And on the notion of trust with the information. that
- 10:01Speaker 1: we're putting in, whether it's personal information or just anything
- 10:04Speaker 1: that we're typing in and receiving information back from these LLMs,
- 10:09Speaker 1: what aspects does blockchain fit in that? And can it
- 10:12Speaker 1: improve the trust in what XYO is working on in
- 10:16Speaker 1: ensuring not just hallucinations, but overall the validity of the
- 10:20Speaker 1: conversation and the information that you're working with on AI?
- 10:24Speaker 3: Well, we just recently announced and we're launching this thing
- 10:27Speaker 3: called CryptoCards, which is a great demonstration app in partnership
- 10:31Speaker 3: with Gate. And Gate has this, Gate.ai, which is their
- 10:35Speaker 3: AI offering along with their crypto stuff. And what we're
- 10:39Speaker 3: doing there is kind of demonstrating exactly what you're talking
- 10:41Speaker 3: about there. Well, how does that integrate and how do
- 10:43Speaker 3: we use cryptographic sovereignty and provenance, especially with X, Y,
- 10:49Speaker 3: O and X, Y, O layer one to basically make
- 10:54Speaker 3: the use of AI a little bit more secure and
- 10:56Speaker 3: a little bit better. And what we do there basically
- 10:58Speaker 3: is we use transparency. So I've always been a fan
- 11:01Speaker 3: of transparency as security where, like for example, a troll
- 11:06Speaker 3: on social media, if they had to go and put
- 11:09Speaker 3: their actual identity next to the thing that they said,
- 11:12Speaker 3: they'd probably think twice about what they said, right? And so,
- 11:15Speaker 3: for example, if we have an AI that's running and
- 11:19Speaker 3: that AI is controlled by a certain entity, if you
- 11:21Speaker 3: have transparency and an audit log of what it did
- 11:24Speaker 3: and who actually did it and controlled that, A, the
- 11:28Speaker 3: repercussions could be more easily followed through on because you
- 11:32Speaker 3: can say, well, we know who did this and we
- 11:33Speaker 3: can go and make them stop or ask them to
- 11:36Speaker 3: stop doing that or whatever it is. Or if it's
- 11:38Speaker 3: a bug, for example, we can go and repair it.
- 11:40Speaker 3: And so really what we do is we focus on
- 11:42Speaker 3: those audit logs. When a game, for example, in crypto
- 11:45Speaker 3: cards happens, what we do is we record all the
- 11:50Speaker 3: provenance of the original hand, which model was used, what
- 11:54Speaker 3: the prompts were for it, and those sorts of things.
- 11:56Speaker 3: We store those in a data lake and then put
- 11:58Speaker 3: those hashes on chain. So that way later on, you
- 12:00Speaker 3: can go look at that and you can review exactly
- 12:02Speaker 3: what happened and potentially see, you know, if you run
- 12:05Speaker 3: that same game with a different model, what would happen
- 12:07Speaker 3: and those sorts of things. So really having that layer
- 12:11Speaker 3: of transparency and transparency the ability to go and see
- 12:14Speaker 3: an audit log of what happened is important. Definitely.
- 12:19Speaker 2: I agree. And I did see that announcement. Congratulations.
- 12:22Speaker 1: I do like Gate and their AI offerings as well.
- 12:25Speaker 1: And I'm excited to see the first hands following along
- 12:29Speaker 1: with that for the game. With that established as a foundation,
- 12:34Speaker 1: how do you see the next steps beyond that in
- 12:37Speaker 1: establishing this trust and validity to more business use cases
- 12:42Speaker 1: outside of crypto as well, but things where there's more
- 12:45Speaker 1: at stake than just a game.
- 12:48Speaker 3: The next steps for applying this to other uses are
- 12:51Speaker 3: things like robotics, for example. What did the robot actually do?
- 12:54Speaker 3: Or anywhere where it matters. And in the game, for example,
- 12:58Speaker 3: the next step there also could be if the game
- 13:01Speaker 3: has high scores and there's a prize for high scores, then, well,
- 13:05Speaker 3: how do we know this was actually done legitimately? Or
- 13:06Speaker 3: how did we know that the house actually did not
- 13:09Speaker 3: put their their hand in after the fact. So did
- 13:12Speaker 3: the person go and have the opportunity to see the
- 13:16Speaker 3: market data in crypto cards prior to playing, or did
- 13:18Speaker 3: they put their hand in prior to the start of
- 13:25Speaker 3: the game? So for example, you have to put a
- 13:27Speaker 3: hash in that's your play for your hands prior to
- 13:30Speaker 3: the game starting, and then the game starts. And so
- 13:33Speaker 3: I can't, you know, at the end of the game say,
- 13:34Speaker 3: oh no, this is the hand that I used that
- 13:37Speaker 3: basically would have won the game and here it is.
- 13:40Speaker 3: Trust me, I didn't cheat. Where if you had to
- 13:42Speaker 3: put the hash in prior to the game starting, the
- 13:45Speaker 3: hand which you reveal has already been locked in basically.
- 13:48Speaker 3: So you can't change it as opposed to won't change it.
- 13:50Speaker 3: So that really gets back to the can't versus won't
- 13:53Speaker 3: question where we really try to add can't to AI
- 13:58Speaker 3: as opposed to won't to AI. Now back to a
- 14:02Speaker 3: point that I hear all the time is people say, well,
- 14:04Speaker 3: there's these settings that says that, you know, AI won't
- 14:06Speaker 3: do this on your computer, or it won't do this
- 14:08Speaker 3: with your data, or, you know, it won't delete your
- 14:11Speaker 3: things or whatever. It's like, well, won't won't is kind
- 14:14Speaker 3: of scary. Can't is a lot better.
- 14:15Speaker 1: Yeah, it would be, uh, well, I, you know, there
- 14:20Speaker 1: are some permissions that you have to allow. Um, but
- 14:23Speaker 1: it's sort of like terms and conditions. People just don't
- 14:25Speaker 1: read it. They just click allow all, and then eventually.
- 14:28Speaker 1: It starts deleting things that you forgot about the terms
- 14:31Speaker 1: and conditions from last year.
- 14:32Speaker 2: Now it can do whatever it wants.
- 14:34Speaker 1: And it keeps enroaching closer and closer to taking over
- 14:39Speaker 1: all of your stuff.
- 14:42Speaker 3: Well, also, it's much more easy to use AI power
- 14:45Speaker 3: if you give it more ability to do that right.
- 14:48Speaker 3: Because what happens is, this happened to me when I
- 14:51Speaker 3: first started using it, where I I want to say
- 14:54Speaker 3: yes to every single prompt, which is yes just once,
- 14:56Speaker 3: yes just once. But then I found myself basically not
- 14:59Speaker 3: really saving much time because all I'm doing is sitting
- 15:01Speaker 3: there reviewing what it does and saying yes or no
- 15:03Speaker 3: and me basically being the guardian. And so for me
- 15:06Speaker 3: to be able to have an agentic experience as opposed
- 15:09Speaker 3: to a prompt by prompt experience, I need to have
- 15:12Speaker 3: enough rights so it can go and do things and
- 15:14Speaker 3: try things and those sorts of things. without me being interrupted,
- 15:18Speaker 3: especially if I have like 10 of these going at
- 15:20Speaker 3: the same time. I just don't have enough bandwidth to
- 15:22Speaker 3: be able to go and review all the prompts for
- 15:24Speaker 3: every single one of those 10 agents. So getting that
- 15:28Speaker 3: right balance of fluidity for your AI usage with security
- 15:35Speaker 3: is difficult. But at the same time, I think transparency
- 15:39Speaker 3: of what actually happened is very important. And they do
- 15:41Speaker 3: keep logs inside of most of these tools that you
- 15:44Speaker 3: can go and look at, which is nice. Because I've
- 15:46Speaker 3: had where one agent complains that something changed in this repo,
- 15:49Speaker 3: for example, because I have a different agent that's doing
- 15:51Speaker 3: something there at the same time. But if they could
- 15:53Speaker 3: more easily communicate and see, okay, well, these are the
- 15:56Speaker 3: things this guy changed, these are the things that those
- 15:57Speaker 3: guys changed, and somehow negotiate what's going on. You can
- 16:00Speaker 3: use work trees, for example, to do that. And GitHub,
- 16:03Speaker 3: to a large degree, is one of those solutions of
- 16:05Speaker 3: how do you actually store things, they hash what's on
- 16:09Speaker 3: GitHub and see what happens. And it has a history
- 16:11Speaker 3: there as well. So GitHub is like a Web2 version
- 16:14Speaker 3: of Providence in many ways.
- 16:18Speaker 2: You mentioned the difference between.
- 16:23Speaker 1: Ensuring the validity of data and then of actions, especially
- 16:27Speaker 1: when there's money at stake. When the game is a
- 16:29Speaker 1: high stakes game and there's money to be earned, or there's,
- 16:33Speaker 1: for example, an AI training program that's out and it's
- 16:38Speaker 1: meant for humans to enter information and they get a reward.
- 16:43Speaker 1: Maybe it's a small reward. Maybe it's, maybe it's big.
- 16:46Speaker 1: How do we prove the authenticity of making sure they
- 16:52Speaker 1: don't cheat? Is there a difference between proving the validity
- 16:55Speaker 1: of data and of actions to prove something was done
- 16:59Speaker 1: at the right time?
- 17:01Speaker 3: Well, the, the goal in most, um, cases is to
- 17:05Speaker 3: try to use the laws of physics to prevent cheating.
- 17:08Speaker 3: So for example, if you have a piece of paper
- 17:11Speaker 3: and you say, well, write your answer on the piece
- 17:12Speaker 3: of paper and then put it face down and I
- 17:14Speaker 3: do the same thing. And then maybe I changed the
- 17:17Speaker 3: answer or maybe I got a separate piece of paper
- 17:19Speaker 3: or it's hard to say where. If we use hashes,
- 17:22Speaker 3: for example, and we say, well, hash your data and
- 17:24Speaker 3: put that on the ledger, That's there forever. I can't,
- 17:29Speaker 3: the laws of physics are that I can't go back
- 17:31Speaker 3: in time and, and change that. And so I've declared
- 17:34Speaker 3: that prior to a certain point of what that answer is. Now,
- 17:37Speaker 3: if it's traditional things to say, well, you write down
- 17:39Speaker 3: your answer. I wrote down my answer. We both give
- 17:41Speaker 3: them to the arbiter of the game. They can go
- 17:44Speaker 3: look at those and they know what the answers are,
- 17:46Speaker 3: but in theory they could be compromised or cheating as well.
- 17:48Speaker 3: And that sort of thing. But the thing which is
- 17:49Speaker 3: nice about a system like X by O layer one
- 17:52Speaker 3: is that we can basically store data. hash in its
- 17:57Speaker 3: provenance without having the data. So what happens is you
- 18:01Speaker 3: hash it, you keep it, you don't give it to anybody.
- 18:04Speaker 3: You give us the hash, we store that. Now, at
- 18:06Speaker 3: the end of the game, you have two choices. Either
- 18:08Speaker 3: you reveal your hand and it matches that hash and
- 18:11Speaker 3: then you play that hand. Or if you want, kind
- 18:14Speaker 3: of like in poker, you can say, well, I'm not
- 18:15Speaker 3: even going to reveal my hand. I'm going to just
- 18:17Speaker 3: muck these cards and throw them away. And at that
- 18:19Speaker 3: point in time, we don't know what your hand was,
- 18:21Speaker 3: but you just you lose by default because you decided
- 18:24Speaker 3: not to reveal. So the one option that's not available
- 18:27Speaker 3: to use, you can't reveal a hand that you didn't play.
- 18:30Speaker 3: So that basically uses time and physics to prevent that.
- 18:34Speaker 3: And so unless you make something which is like physically impossible,
- 18:37Speaker 3: it's kind of like a zero knowledge proof. A zero
- 18:39Speaker 3: knowledge proof just uses math to make it so that
- 18:41Speaker 3: it's perfect. And X, Y, O layer one basically uses
- 18:45Speaker 3: time in many cases to store information about information it
- 18:50Speaker 3: doesn't even have. And so that's what makes it impossible
- 18:54Speaker 3: for a person to cheat there. So even if I
- 18:56Speaker 3: have a log, let's say, for example, I have an
- 18:57Speaker 3: audit log, I can have 10 gigabytes of audit log
- 19:00Speaker 3: that's in a database or a data lake in our case.
- 19:03Speaker 3: If I put that hash in there, Unless I provide
- 19:05Speaker 3: that entire log, I haven't provided the log. So a
- 19:09Speaker 3: person could say, well, I deleted the log and I'm
- 19:11Speaker 3: not going to provide it to you. But then we
- 19:12Speaker 3: assume that they cheated. It's kind of like the IRS
- 19:15Speaker 3: coming to you and asking for your receipts. And you're like, well,
- 19:17Speaker 3: I don't have any receipts at all. It's like, well,
- 19:19Speaker 3: that's a problem for you. But if you can prove
- 19:20Speaker 3: the receipts, then they're there. Say, for example, in tax filing,
- 19:26Speaker 3: if I could hash all my receipts, send that hash
- 19:30Speaker 3: to the IRS. And then when they come and ask
- 19:32Speaker 3: me for my receipts, I can prove that the receipts
- 19:34Speaker 3: I use are this hash that would carry more weight
- 19:37Speaker 3: than it's like, Hmm, maybe I, you know, I had
- 19:40Speaker 3: these receipts or I didn't have those receipts. So it's,
- 19:45Speaker 3: important to use things which humans can't affect to make
- 19:49Speaker 3: a can't as opposed to a won't. Definitely.
- 19:51Speaker 1: And in this case, it's a beautiful use case for
- 19:55Speaker 1: blockchain in that it is that piece of paper you
- 19:58Speaker 1: write on before, but you can't change it.
- 20:00Speaker 2: You can't swap it out.
- 20:02Speaker 1: And physics and time really puts blockchain in the perfect
- 20:08Speaker 1: place to support AI. But you said in the past
- 20:12Speaker 1: that not everything should be on blockchain, and that's what
- 20:14Speaker 1: the use case of the hash is.
- 20:17Speaker 2: Could you.
- 20:19Speaker 1: Explain the difference in what is important? Is it just
- 20:23Speaker 1: the hash of the information.
- 20:24Speaker 2: That needs to be on the blockchain?
- 20:25Speaker 1: And what else should be on the blockchain and what
- 20:28Speaker 1: definitely shouldn't be?
- 20:31Speaker 3: Well, it's definitely important to minimize the size of the
- 20:33Speaker 3: blockchain because one of the problems that Ethereum has is
- 20:35Speaker 3: everything is on the blockchain. So you end up with
- 20:37Speaker 3: a very, very large shared hard drive base that everybody has.
- 20:41Speaker 3: Every blockchain has that to some degree, but you want
- 20:43Speaker 3: to minimize what those are. We call them elevated payloads
- 20:46Speaker 3: where the elevated payloads actually go on chain and they
- 20:48Speaker 3: have to be there and they get validated. But then
- 20:50Speaker 3: the non-elevated payloads are things that are in data lakes.
- 20:52Speaker 3: So for example, if I want to take a movie
- 20:54Speaker 3: and hash that movie, I can store that in the
- 20:56Speaker 3: data lake or I can store it somewhere else and
- 20:58Speaker 3: put the hash on chain. The downside of that is
- 21:02Speaker 3: it's loss is losable as far as I can, I
- 21:04Speaker 3: can lose that, that data. And so people, when they
- 21:08Speaker 3: hear the word permanence, for example, we talk about, um,
- 21:10Speaker 3: ex bio layer one provides permanence. There's two definitions of
- 21:14Speaker 3: permanence that I think are important to differentiate. One is
- 21:18Speaker 3: it can't be lost, which is often what people think
- 21:20Speaker 3: of permanence. And one is it can't be changed. Refocus
- 21:24Speaker 3: on the second one. It can't be changed. So I
- 21:27Speaker 3: permanently locked in my hand that I played, for example,
- 21:30Speaker 3: is permanence. I permanently put my hand on Ethereum's blockchain,
- 21:35Speaker 3: so I can't lose it is also permanence, but the
- 21:39Speaker 3: different permanence. So the permanence of I can't lose it
- 21:41Speaker 3: is very expensive. The permanence of I can't change it
- 21:45Speaker 3: can be very inexpensive. It's just one hash basically for
- 21:47Speaker 3: as much data as you want to. So we really
- 21:50Speaker 3: try to tailor X by layer one to be as
- 21:54Speaker 3: optimal as possible, which does make it much better for
- 21:58Speaker 3: certain use cases and also makes it not possible necessarily
- 22:00Speaker 3: for other use cases. So if you have to have
- 22:02Speaker 3: your data not losable, that's something which is not really
- 22:06Speaker 3: practical for X, Y, O layer one. But I would
- 22:08Speaker 3: also argue that it's not really practical for anything that's
- 22:10Speaker 3: out there because things can still get lost. Definitely.
- 22:15Speaker 1: Yeah. especially those private keys from Bitcoin back in 2010.
- 22:20Speaker 1: I wish we had those. What are your thoughts on
- 22:25Speaker 1: how much focus there is on the synergy of AI
- 22:29Speaker 1: and blockchain throughout the blockchain industry? I know there's a
- 22:34Speaker 1: couple of protocols that are focused on decentralized AI and
- 22:38Speaker 1: improving AI. I feel like a lot of the focus
- 22:41Speaker 1: is on finance, which is important, of course, and an
- 22:43Speaker 1: early use case.
- 22:44Speaker 2: But maybe there's not.
- 22:45Speaker 1: Enough discussion going on right now in understanding how important
- 22:50Speaker 1: blockchain is to building up AI properly.
- 22:54Speaker 3: Well, I think it's kind of an interesting situation. This
- 22:57Speaker 3: comes along every few years where people are like, oh,
- 23:00Speaker 3: AI is a hard word and blockchain is a hard word.
- 23:03Speaker 3: Let's put them together and put them in a pitch
- 23:04Speaker 3: deck and see what happens, right? So I think often
- 23:07Speaker 3: you'll see more projects that are trying to merge them.
- 23:12Speaker 3: And in some cases, it's just like have a AI
- 23:16Speaker 3: bridge or some sort of AI router, for example, that
- 23:18Speaker 3: uses blockchain. And it's, well, how is it better with
- 23:21Speaker 3: blockchain in that case? So you always have to ask,
- 23:23Speaker 3: you know, is it better if you combine these two
- 23:26Speaker 3: things or is it not better? And I'd say in
- 23:29Speaker 3: most cases, it may not be better necessarily. It's just fuzzy.
- 23:33Speaker 3: It's not better. But in many cases, like in that
- 23:37Speaker 3: layer one case, I think there's a distinct usage for
- 23:40Speaker 3: it and a way to make the system better. where
- 23:44Speaker 3: if I could have a provable history of say robots
- 23:48Speaker 3: or autonomous hardware that's running around, especially if they can
- 23:53Speaker 3: check in with each other, cosign things and then put
- 23:56Speaker 3: those on the blockchain, I can see where these items were.
- 23:59Speaker 3: And in a weird sort of way, it's almost like
- 24:01Speaker 3: a real life version of our coin app where having
- 24:04Speaker 3: these things check in autonomously into the blockchain allows you
- 24:10Speaker 3: to see what actually happened in the real world, because
- 24:13Speaker 3: I think it's going to be one of the, one
- 24:14Speaker 3: of the difficulties is going to be, well, you know,
- 24:17Speaker 3: something bad happened. Um, was it my robot that did it?
- 24:20Speaker 3: Was it your robot that did it? Was it, yeah,
- 24:22Speaker 3: it was on purpose. Was it by accident? And if,
- 24:24Speaker 3: you know, having audit logs for those things that are
- 24:26Speaker 3: constantly being stored and also constantly being, um, like bookmarked
- 24:30Speaker 3: on an X by layer one system, we'll make it
- 24:33Speaker 3: so that we can go back in time and we
- 24:35Speaker 3: can at least, um, find some sort of direction for
- 24:38Speaker 3: where it is. And so, an actual solution for an
- 24:41Speaker 3: actual problem is important to have. And so I think
- 24:46Speaker 3: some cases where we see that synergy, that's not the case.
- 24:51Speaker 3: In some cases, it's extremely true to be the case.
- 24:54Speaker 3: And so I think there's probably more place for blockchain
- 25:00Speaker 3: in AI or in the world now with AI around
- 25:03Speaker 3: than there was before, just because it's a natural fit.
- 25:08Speaker 2: I agree.
- 25:08Speaker 1: I'm looking forward to seeing it more And I think
- 25:10Speaker 1: another major synergy we haven't spoken about yet is not
- 25:15Speaker 1: just AI and the data provenance of blockchain, but the
- 25:19Speaker 1: payments and the wallets of AI agents or humanoid robots
- 25:24Speaker 1: or self-driving cars and things like that. Is there a
- 25:27Speaker 1: plan for XYO to extend beyond data to ensuring the
- 25:34Speaker 1: validity of payments or the success in having that infrastructure expanded?
- 25:38Speaker 1: in there as robots start interacting and paying on our
- 25:41Speaker 1: behalf or paying each other?
- 25:42Speaker 3: Well, we're looking more at facilitating payments as opposed to
- 25:47Speaker 3: validating other people's payments. So for example, one of the
- 25:50Speaker 3: things which we built into CryptoCards that's not really that
- 25:53Speaker 3: easy to see unless you go and look at the
- 25:55Speaker 3: blockchain and understand it is we have an eventing system
- 25:58Speaker 3: where basically on our blockchain, a person can go and
- 26:01Speaker 3: register for an event by assigning a payload with their address.
- 26:06Speaker 3: And then the eventing server will start sending those events
- 26:08Speaker 3: to you through like a webhook, for example. And then
- 26:14Speaker 3: it can stop when you stop paying for that. But
- 26:15Speaker 3: what you can do is you can say, well, I
- 26:16Speaker 3: want to get this event for the next, say, 1,000 blocks.
- 26:20Speaker 3: And I'm going to pay 3XL1 to get those events
- 26:22Speaker 3: or a certain amount per event. So the actual payment
- 26:26Speaker 3: of microservices, I mean like micro-microservices, like nano-services, is something
- 26:31Speaker 3: which I think is going to become much more prevalent
- 26:33Speaker 3: where A robot or AI might be like, hey, I
- 26:38Speaker 3: want to know for the next two minutes, when does
- 26:42Speaker 3: a new block appear on X by layer one? And
- 26:45Speaker 3: I want to get those events pretty frequently for the
- 26:47Speaker 3: next two minutes and then stop. But I don't want
- 26:49Speaker 3: to go and have to create an account with a
- 26:52Speaker 3: credit card and that sort of thing to be able
- 26:54Speaker 3: to use the API for whatever it is. And so
- 26:57Speaker 3: I think real-time ad hoc usage of services that get
- 27:01Speaker 3: paid for via blockchain payments in a very, very efficient
- 27:04Speaker 3: way is going to be extremely important for AI and
- 27:07Speaker 3: for robotics because of the fact that they need a
- 27:10Speaker 3: very broad spectrum of access to things. And they need
- 27:15Speaker 3: to be able to pay for those things efficiently and
- 27:16Speaker 3: to authenticate for them efficiently without having to go through
- 27:19Speaker 3: all the pomp and ceremony of credit cards and accounts
- 27:23Speaker 3: and logins. Definitely.
- 27:25Speaker 2: And micropayments has been a.
- 27:28Speaker 1: Supposed use case of blockchain for many, many years, but
- 27:32Speaker 1: we haven't really seen it. at least in real world payments.
- 27:35Speaker 1: But now with AI and token usage, we're starting to see.
- 27:39Speaker 2: Every word is a micro cent.
- 27:42Speaker 1: And that seems like the perfect use case for micropayments.
- 27:46Speaker 1: We're still putting your credit card into the AI and
- 27:49Speaker 1: just preloading it. But I feel like there's a perfect
- 27:52Speaker 1: use case for micropayments on blockchain with AI. It's like
- 27:56Speaker 1: the number one use case that's now coming to light.
- 28:00Speaker 3: I thought it was actually kind of funny when they
- 28:02Speaker 3: came out with the AI thing and called it tokens.
- 28:04Speaker 3: It's like, how many tokens in, how many tokens out?
- 28:06Speaker 3: I'm like, well, I've heard the word tokens before. It's
- 28:08Speaker 3: an ERC-20 token. So they're kind of modeling that to
- 28:12Speaker 3: some degree where it's a token economy. I'm surprised they
- 28:17Speaker 3: haven't said yet, well, I'll pay you in 500 Claude
- 28:22Speaker 3: tokens and I'll use that as money to pay you
- 28:23Speaker 3: with or something like that. But there's no way for
- 28:25Speaker 3: me to actually transfer tokens from one account to another
- 28:28Speaker 3: account on Claude before they know. But there's nothing saying
- 28:31Speaker 3: that I can't use their tokens as a payment method.
- 28:36Speaker 3: But I think definitely using blockchain as a way to
- 28:40Speaker 3: autonomously give tokens, for example, to an agent for its use.
- 28:46Speaker 3: is something which we're going to be doing very often
- 28:49Speaker 3: in the near future, as opposed to saying, well, here's
- 28:51Speaker 3: my login for my cloud account, use as many tokens
- 28:54Speaker 3: as you want. That's a little scary, where if I
- 28:56Speaker 3: can give it an allowance with a blockchain system, it's
- 28:59Speaker 3: something which I think people would definitely want to use.
- 29:00Speaker 3: But the token economy is something which I think will
- 29:05Speaker 3: blossom a lot under AI, just because AI is so
- 29:08Speaker 3: much more fragmented in the way that it uses the world,
- 29:10Speaker 3: as opposed to a human where it's like, well, I
- 29:12Speaker 3: don't want to do a microtransaction for every single bite
- 29:15Speaker 3: of food that I eat at a restaurant, for example,
- 29:17Speaker 3: or for every minute of the movie that I watch.
- 29:20Speaker 3: I want to just get the whole movie. And so
- 29:22Speaker 3: microtransactions have kind of, I think, failed to this point
- 29:26Speaker 3: with blockchain for those two reasons. One, it's just the
- 29:29Speaker 3: gas cost for Bitcoin and for Ethereum is way too expensive.
- 29:32Speaker 3: But two, it's humans don't think in those micro steps
- 29:37Speaker 3: the way that AI can.
- 29:39Speaker 1: Yeah, no, that makes perfect sense. And, you know, it's
- 29:42Speaker 1: two months since our last video. I would love to
- 29:44Speaker 1: get you back on once more at the end of
- 29:46Speaker 1: the year and see what Q4 has in store for
- 29:49Speaker 1: the growth of AI. It's just getting faster and faster,
- 29:52Speaker 1: as well as blockchain now kicking back up, not just
- 29:56Speaker 1: the development of blockchains and you know, X, Y, O
- 29:59Speaker 1: layer one is developing super fast.
- 30:00Speaker 2: I understand using AI, but also the prices are going
- 30:03Speaker 2: back up.
- 30:03Speaker 1: So people are getting more interested and more people are
- 30:06Speaker 1: looking at different ways that these two technologies can synergize together.
- 30:11Speaker 1: What is some of the most important things that are
- 30:14Speaker 1: happening at X, Y, O between now and the end
- 30:16Speaker 1: of the year that we may have to look forward to?
- 30:19Speaker 3: Well, it kind of says something interesting there, which I
- 30:21Speaker 3: think is very appropriate for thinking about AI in the
- 30:25Speaker 3: context of X, Y, O layer one. Well, we use
- 30:27Speaker 3: AI in two different ways. One is we use AI
- 30:29Speaker 3: to develop and to accelerate how fast we can do
- 30:32Speaker 3: really cool things and make new products with X, Y,
- 30:35Speaker 3: O layer one. So AI as a tool for us
- 30:37Speaker 3: has been fantastic. And we've really dove into that, made
- 30:41Speaker 3: a bunch of skills and that we use ourselves, we
- 30:44Speaker 3: share with our partners and they can use the skills
- 30:45Speaker 3: as well. And then also we prioritize AI things. Kind
- 30:49Speaker 3: of the crypto cards is that where the house, which
- 30:51Speaker 3: plays the hand of crypto cards uses AI the AI
- 30:55Speaker 3: from gate, for example, in this case, to actually decide, well,
- 30:58Speaker 3: how should I play that? And so prioritizing AI is
- 31:00Speaker 3: the other aspect of it. Well, how do you use
- 31:02Speaker 3: AI as a feature in the product that you want
- 31:05Speaker 3: to use? And so we're really focusing on both of those.
- 31:08Speaker 3: We're focusing on using AI as a tool to accelerate
- 31:10Speaker 3: ourselves and reusing AI as a feature that we can
- 31:14Speaker 3: use in our products and in different things and interlacing
- 31:17Speaker 3: those features with hashes and with provenance and sovereignty through X, Y,
- 31:23Speaker 3: or layer one.
- 31:25Speaker 2: I'm definitely going to test out these crypto card first hands.
- 31:28Speaker 2: I saw the announcement.
- 31:29Speaker 1: I'm subscribed to when the first hands go live. I'll
- 31:32Speaker 1: put it in the show notes below because we've mentioned
- 31:34Speaker 1: it quite a bit. I appreciate your insights into everything AI.
- 31:38Speaker 1: You're such a smart brain. And I'm looking forward to
- 31:42Speaker 1: talking with you again in the near future. Thank you
- 31:44Speaker 1: so much for the time, Ari.
- 31:46Speaker 3: I appreciate it. Thanks so much for having me on.