[00:00] You don't know?
[00:01] That's equivalent to the status status.
[00:04] So attention is all-in.
[00:06] It's very interesting.
[00:07] It's very fun.
[00:08] So to make an LNN, you have to understand all the steps
[00:12] like the corpus, and then bring to the attention.
[00:16] And the transformer is a key that Google actually invented.
[00:20] It's really, really invention that you can do the
[00:24] power processing in your GPU.
[00:25] That's the reason why it's drastically, actually, got
[00:27] smarter.
[00:28] You can put any vectors into any GPUs.
[00:32] And then do a lot of actually cheat after that.
[00:37] There's a lot of cheat in LNN.
[00:39] And then do the outputs with the input of the
[00:44] prompts to predict.
[00:46] It's just a good prediction of the probability of the
[00:50] next constant based upon the previous tokens.
[00:55] But now I'm so excited about learning about LNN, and I want
[01:01] to make something creative.
[01:03] It's probably difficult, but I'm trying.
[01:06] Other than that, I'm making the systems in the BFIRE,
[01:10] spending $3,000 worth of tokens a day.
[01:17] So my mission for now is to decrease the tokens cost.
[01:22] $3,000 a day is too expensive, even for BFIRE.
[01:26] So thinking about the local LNN, or maybe I'm making
[01:33] the structure to distribute the prompts into the
[01:35] different AI, and the different models based upon
[01:38] the situations.
[01:40] That's awesome.
[01:43] One of the things I'm also curious about is how do you
[01:47] think stable coins will change consumer behavior?
[01:51] Japan has some payment options for consumer, but if
[01:55] everything cannot be represented on a chain, do you
[01:57] feel like most of how we interact and pay for things
[02:00] stays the same?
[02:01] Does things become better?
[02:03] What would that look like in a few years?
[02:04] Yeah, we have a 50 different payment systems.
[02:07] You need to choose all the time.
[02:10] I really want all the payment guys consolidated into
[02:13] maybe two, but PayPay, whatever, Suica, that's too
[02:19] many, seriously.
[02:22] But if you have the remaining time, we'll call it.
[02:24] Stable coins, 450, I think the US is not different.
[02:31] I think it's more like a merchant side.
[02:33] Merchant actually pays maybe 2% to 8% to the issue of
[02:38] the payment, issue of the Suica, PayPay, or credit
[02:44] card, Visa, Master, and CCB.
[02:46] Maximum 8% is really, really huge, and it's
[02:49] for the merchant.
[02:49] If that has gone down to maybe very close to 0% or 0.1%,
[02:54] the merchant may give that extra credit
[02:57] or points to the users.
[02:59] So that may be a difference.
[03:01] Otherwise, you are UX, and then I think stable coins
[03:04] pretty much same.
[03:07] Yeah, and it's just mostly making it easier for the
[03:10] consumers to have loyalty.
[03:12] Yeah, I think so.
[03:13] But the question here is how all the shops in Japan,
[03:17] even the local or rural people, very elder people,
[03:24] can do the stable coin.
[03:26] That's the question.
[03:26] And how do you sell?
[03:28] How do you do the sales to the people?
[03:32] And you don't pay pay, so think, spend a billion dollars
[03:35] and a billion dollars to expand the channels, not in
[03:41] Tokyo, but also all the regions of Japan.
[03:44] And stable coin, a lot of people, it's too optimistic
[03:46] that people will try to download the stable coin
[03:50] wallet and automatically do the payment.
[03:52] That never happens.
[03:54] So who's going to do it?
[03:55] That's the question for me.
[03:56] Which means there's an opportunity for somebody to
[03:59] make that easier.
[04:01] Couple of more questions.
[04:02] So you also have a podcast, and you've talked about
[04:05] interesting ideas there.
[04:06] And one of the recent episodes was about Ethereum.
[04:09] And you posed a question about what does it mean for
[04:13] Ethereum ecosystem to succeed, and how does that
[04:15] look like, relative to the Ethereum token?
[04:19] I'd love to hear your thoughts on how you think that
[04:21] looks like, and what do you think the broader
[04:24] Ethereum ecosystem could look forward to?
[04:27] Sure.
[04:29] In my career, there's only two excitements.
[04:33] Bitcoin and Ethereum.
[04:35] LMs.
[04:37] LMs.
[04:38] And otherwise, other blockchain is kind of copycat for me.
[04:48] So the Ethereum has a huge negative effect.
[04:53] Although, it started in 2012, maybe?
[04:58] 2012 is right?
[04:59] Vitalik, actually.
[05:00] 13, yeah.
[05:02] Yeah.
[05:02] And then I joined your first, actually, token sale.
[05:08] It was actually issued around $5, $10, $11.
[05:14] I think I bought maybe $10?
[05:16] It was $0.20, I think.
[05:18] $0.20?
[05:18] Then I did join the first class.
[05:21] I joined probably around $5 to $10.
[05:27] It's a huge negative effect.
[05:29] So L2, do you think L1 or L2, a lot of people,
[05:35] there's a huge discussion on L2.
[05:38] But then now, do you really need L2 anymore?
[05:44] And then there's a hack in the L2.
[05:48] It's a sidechain.
[05:50] If the L2 chain is not secure, that's going to be a huge trouble.
[05:59] But the Ethereum L1 is very stable.
[06:02] Very stable, a lot of people actually maintenance,
[06:04] and do forecasts, and a huge, huge negative effect.
[06:08] And then that's a base.
[06:10] And then you can build a lot of things on the top of it.
[06:13] And then people don't want to realize that's kind of OS,
[06:16] right?
[06:16] VM, virtual machines kind of OS.
[06:19] So now, it's more important to have a very stable,
[06:24] tamper-resistant blockchain, back-and-back, very technical,
[06:28] new technology.
[06:30] And this is very, very strong.
[06:31] I think it's a very stable base.
[06:34] Yeah, so, yeah.
[06:36] Does that mean it's interesting to more than just developers?
[06:41] Because the network effect is good to have things deployed.
[06:44] But how do you think that?
[06:45] I think for the purpose, it should
[06:47] choose the stable blockchain.
[06:50] Because if you worry about the hacking,
[06:53] how many actually worksheets hacked so far?
[06:55] One a week.
[06:57] One week.
[06:58] Better to choose a stable one.
[07:00] Having lots of multiple functions.
[07:03] You can build it on top of it.
[07:05] That makes a lot of sense.
[07:06] And my last question to you is, for anybody in this room,
[07:11] especially somebody who is in the Japanese market trying
[07:14] to build something in crypto, what advice would you give
[07:17] them?
[07:17] And where would you say the opportunity is?
[07:19] I think something like related to AI is very challenging.
[07:22] And I'm very sensitive to making AI-specific blockchain.
[07:28] I want to change the AI only AI can access to you.
[07:34] That's easy.
[07:35] You do the puzzle to prove you're a human being,
[07:40] to log in, right?
[07:41] So to prove you are an AI, it's very easy.
[07:44] You can ask a very difficult question.
[07:46] You cannot solve it in a second.
[07:50] Right.
[07:51] And then only giving the permission to the AI
[07:55] is very, very useful.
[07:58] Why?
[07:59] Because you can skip the regressions.
[08:01] A lot of regressions is designed
[08:05] because the human being is actually using it.
[08:07] If the AI pays the money to other AI,
[08:11] do you think this is under the payment act law?
[08:15] Maybe not.
[08:16] We don't know.
[08:17] That's great.
[08:24] If you make a bank only for AI,
[08:26] using a unique bank license,
[08:28] we don't know.
[08:29] Because a lot of regressions,
[08:31] not only Japan but also the US,
[08:33] is designed for the human being.
[08:36] To follow the rules.
[08:37] So we don't know.
[08:39] That's a really good point.
[08:41] Maybe a bank should just be a smart contract too.
[08:46] Well, thank you so much for the amazing conversation.
[08:48] Please give a big round of applause.
[08:50] Thank you very much.