Transcript
Speaker: Welcome to a new episode of the Insilico Terminal podcast. My guest today is another anonymous anime profile picture. I just yesterday realized that it's Makima, right?
Speaker: Yes. which i so yes i didn't I didn't properly like look at your PFP before, but then yesterday I saw the frame and I realized it's sad. But yeah, my my my guest today is IRP pairs, but I think you actually don't do that much. Like, how how did your name come about? sarah called i our pers ah so So my name started because that was the first thing I ever did when I was trading.
Speaker: You know, i was really interested in crypto. And i when I finally decided to sit down and start trading, I was like, okay. I need to do something that makes sense. like I have to understand why I'm making money. And then I thought, well, if I can if i can take two different pairs and just like actually trade them in some way where there's a price disconnect, this makes sense.
Speaker: And the the first trades that I did were that. They were like very simple on-chain arbitrage. And then I made this account. And then ah the Makima picture happened because... like I mean, i I watched Chainsaw Man and I read the manga and I was never even that much into it, but...
Speaker: I liked that picture because it's the one where she's like framing everything and controls it. I thought, oh, that's perfect because like I'm reading all the information on chain, so I'm going to make this my picture. That's funny.
Speaker: what what Do you remember what you what the first who where the first pairs were when traded? Yeah, the the the most significant like trade where I would say like this is the first main one is I did an ARB on a privacy token called Targhee.
Speaker: um This is like a total, the project doesn't matter anymore. I think they like rugged the treasury or whatever. like it was a complete you know failure. But at the time it was pretty interesting to some people because they were doing this ah like Monero side chain.
Speaker: And yeah they had a it was proof of work mining system. And they had ah a pre-mine and everything. ah like Not a pre-mine. It was like ah um you could do the testing in the alpha and they would give you an airdrop. That's what it was. And so I was aware of this before it came out. And when it did drop, the only way that you could use the chain was through their like GUI application.
Speaker: And then they said... oh we have we have this uh we're gonna do a wormhole bridge like layer zero i think is what they did it on but um they were gonna bridge the tokens to ethereum so you could wrap your your tokens and the token at the time was xtm because there was a layer one and a layer two but like at the time it was just xtm and so you could wrap it in in their bridge and it would be wxtm and when they did this um There were some big privacy nut guys that had just mined a shit ton of these tokens. And a few of them were like, I'm going to custody this on Ethereum. So they bridged over this like huge amount of Atari. And there was a Uniswap pool with like 1 million USDT and WXTM.
Speaker: And I knew immediately that this was going to be like an opportunity because you could only use the bridge inside their app and it was really janky. And it was a one way bridge, which like if you know anything about cross chain stuff, that's crazy. So like the only you'd send XTM in and it would become WXTM and you couldn't reverse it like you had to sell to USDT.
Speaker: And there was only one exchange that you could buy XTM on, and it was called SafeTrade. And it looks kind of sketchy. And everybody who's like an old head in this space has told me, oh, like SafeTrade is really good. and And to their credit, like they processed all my stuff. But I remember i made ah i made a Twitter thread about this, and they actually got really mad at me. They went to my DM, or they went in my mentions. Yeah.
Speaker: They're like, we're not a scam. We're not sketchy. Like ask anybody. I'm sorry. I'm sorry. But... Yeah, so you you could buy regular XTM on there. And so I was like, holy shit, this is going to be amazing. And right off the bat, there was a 30% disconnect. Like the price in the Uniswap pool was close to, I don't remember if it was 10 cents or a dollar. Like it was it was like 30% more than what it was on the main exchange.
Speaker: And I went, oh, this is amazing. And I was broke at the time. I had no money in crypto. So I took like a cash advance on my credit card. And yeah i bought usd i bought a USDC or USDT, whatever you needed to buy it on SafeTrade. And I bought the Atari and I bridged it out to my wallet. And you had to wait. And I'm pretty sure it's like some guy like manually approving it. You know mean? So I get it finally in my wallet and I take it into the app and I bridge it. And I got all the way through and I made like 25%. I'm like, holy shit, like this is great. And so i kept doing it.
Speaker: And it was really hard too because the bridge had issues at the beginning. Like you could only really use it through the GUI and it was a cross-chain thing. So it was difficult to automate. And I thought this was before AI was getting really good. It was still like kind of janky. When when when was this? Like around which time? Let me see. It would have been... see.
Speaker: Atari launch date.
Speaker: Mainnet was May 25th. Okay. So last year. Yeah. Yeah. like yeah i was i've never I've never heard of any of this stuff. I've never heard it. Beyond reading your Fed, I mean, I read your Fed and shit, but I've never heard of the coin or this this exchange or whatever. Well, this is what I'm saying. There's so many opportunities like this. You know you just have to find them. And yeah it it got really... It was hard, too, because I didn't want to automate it because I was worried it would go away.
Speaker: And it wasn't so easy to do where you could just like use $200 ChachiBoutie sub or whatever and just instantly make what you needed. And there was no documentation because it was brand new. And there were glitches. Like if you tried to bridge your entire amount of Tari or like whatever the token was, it would just fail, but it wouldn't fail right away. it would just get hung for hours. So I had to figure out stuff like how to get it through faster. And I was like constantly rebooting the application and trying like, you know, not the rounding on the token was wrong and like making sure it would go through.
Speaker: And there was me and like two other guys and I was using um one of those on-chain tracker things like when people trade shit coins and you can see like certain wallets repeatedly trading, you know, memes and stuff. And there was like two other addresses that were also doing this. And I was like, oh shit, like I need to really get on this. And I thought it would go away. And what happened was the price of Tari started to decline over time as well.
Speaker: because they weren't really doing anything. And so the hype was fading, but that meant there was always a disconnect between safe trade and the wrapped ah bridge pool. And so I did this for like three days. it was like literally three days straight. And the whole trade was just like click, click, click. And you had to have the nuts to like send it to this exchange you didn't know anything about. And I didn't want ask anybody about it either because I'm like, what all takes is one guy with like a hundred grand and he's going to kill this in like five minutes. Like it will be over.
Speaker: But that was a great trade. It worked. It did exactly what I thought it was going to do. And I made money. How much capacity was there in there? like Well, there was a million dollars in the in the Uniswap pool and there was a 30% price disconnect.
Speaker: So assuming that you could get the Tari tokens, I mean, you could have made like... low six figures i'm not sure exactly um if if somebody had been mining though they could have easily done it like if somebody was willing to rent enough gpu but i just really didn't think it was going to take that long and i also didn't have them i didn't have that kind of money to go spend like 50 grand on gpus or something for this garbage coin and so have you only been in crypto since like last year
Speaker: Yeah, it was like late. when did the AI like Solana coins take off? You know, like Goat and Peanut and like that era? I think end of 24?
Speaker: That would have been when I got started. oh like like a roll I think around the election or like a little bit after, I feel. Yeah, um I would have started late, late 24. I did an internship with a guy who was running a Web3 VC yeah studio. And that was actually where I had heard about this project because it took a while for Tari to launch.
Speaker: Yeah. And then when it finally did, I was like, hey, I wonder what's going on with that project. And I went, holy shit. I went and did that. I was i was just going to ask how you go about like finding stuff like this. so it's really It sounds very obscure.
Speaker: Yeah, I read everything. and And if you go back at like my very early posts, um I had a whole different style and everything. I was still ah capitalizing things like punctuation, like trying to make good content for people. And I remember.
Speaker: Yeah. And early on, I said that I was like, read all of the documentation. and i still think that that's true you know the landscape is changing and like it's even compared to 24 25 like we're in a different place now but i still think actually understanding what you're trading is the best edge that you can have because crypto is still not really fully regulated and so there's just not a consistent expectation of what you're dealing with and i think being willing to like put that time in and take the risk too because there is a lot of tail risk that comes along with touching some of that weird stuff yeah
Speaker: did you Do you have any other trades that are like similar or different categories or whatever that also are good examples maybe that you can like talk about and explore, I guess? Yeah. Well, more recently, um i made pretty good money trading oil when the war started. And that started off... ah That was like ah a total retard trade because I knew a guy who lived in Bahrain and he called me on Telegram in the middle of the night and he's like, they're bombing Bahrain. like And i was like, okay, well, i I have an idea of what we should do. So we put a trade on um and because I had started trading, like I don't know if you remember or not, but
Speaker: when the war started, like oil, the oil markets had almost no volume and then they became like one of the biggest markets. Like it was bigger than Bitcoin for a little bit in terms of daily volume. On hyperliquid you mean? Yes, yes, on hyperliquid. Yeah.
Speaker: um And so i I started looking at that instrument. I was like, well, how does this work? Because I don't really have a traditional finance background. And that was more or less because I didn't go to like target schools, didn't think that I could break into that industry. So crypto always seemed more accessible.
Speaker: And I looked at it and I read about how 24-5 markets work and just like commodities trading in general. and the oil markets had just a ton of rules attached to them and i mean all of those markets did but like they had the weekend trading was different and the after hours trading was different and we were reading about that and i thought man like this is not going to be a system that people know how to deal with and early on they had very narrow bumpers it was like a lower and upper price bound and When the markets closed for the weekend, it went to some kind of weird internal Oracle system.
Speaker: And in the very beginning, it was just like plus or minus 5% or something like that. Like it was totally static. And the war popped off. And it was like over the weekend. And I went, OK, this is going to gap up.
Speaker: And it was pegged. And so we just sat there. We're like, people are going to get ADL. This is what's going to happen. Like the market's going to open and it's going crazy. And that's exactly what happened. But then the next weekend, we got to the next weekend and the rules were the same.
Speaker: And it happened again. And we did it again. I was like, this is stupid. Like, there's no way we can keep doing this. And then they started changing the rules. They started having like the EMA internal oracle thing where it would move and they were changing after hours versus weekends. And long story short, the the trade switched from just being on the right side and ADL to ah people were overreacting and they were directionally correct on which way oil would move. But they didn't understand that in traditional markets, things can only move so fast. There's like breakers that kick in and stuff like that. And we went, oh, well, the Oracle is going to turn on. This thing is going to say, actually, we're not at you know this price. We're down two dollars from here for like one second. And then everybody's going to get wiped out.
Speaker: And so we did that. We did it the opposite direction. And we literally took trades like five minutes before market opened, made 10% or 20%, and then get out. And we we did that for a while.
Speaker: And then there was another thing on Hyperliquid with the oil where Felix had their own market. i don't know if you remember, they had the USDH version of oil. Well... Trade XYZ versus Felix, one of them had, i can't remember which, I think it was Felix, had 0% funding after hours. It went flat, and the other one still had funding. And we were like, what the fuck? like We can just sit here and just eat the funding, and we can make funding over the whole weekend. And it was literally like five or six weeks ago this, and...
Speaker: I think the reason why that worked so well is it was an imperfect market because they were targeting like contracts that roll. So they had these like weird systems and the Oracle wasn't super predictable and they kept changing the rules. And the other thing was there were some postmortems on this about Twitter. I noticed later and I thought about it and I think this makes a lot of sense. A lot of the sophisticated ah traders and like firms and stuff that are in funds that are actually going in and trying to make the market in these perps. didn't have a good way to hedge their position, like at all.
Speaker: It was a very strange situation. And so I think we were just nimble enough where we could sit there at like 2.30 in the morning, read the documentation and go in. And we're also not managing like millions of dollars in capital, you know, at this point. So we're like, well, if we lose, we lose, but we're probably not going to. And this just keeps making money. So let's keep trading it. And then the smartest thing,
Speaker: that I did was walk away. Like after about a month and a half, they had come up with these rules where the price balance would actually move after a certain period of time. It was like a volatility dampener and they had everything all figured out and everybody was paying attention to oil and I was like, I'm done.
Speaker: And the government started intervening. the US s government started trying to like push oil back down. i'm like, I don't want to touch this anymore. Like we're going to get ran over.
Speaker: I was just going to ask how long this persisted, but I guess stepping away at that point is actually really smart. And there's more attention on it and the rules have adjusted a bit more to like where you can't fuck with the contracts that much anymore.
Speaker: yeah Yeah, and that had really good capacity as well because you could get access to good leverage. There were multiple markets. like Even Lighter had a market. We didn't really use it much because it was it was such a it was so great that you could get access to Felix's market as well because you got another oil market with different rules denominated in a different stablecoin.
Speaker: So it was so great to be able to go in. There was like all kinds of pair stuff that you could do with that. And the Felix market was also... less liquid so there was a ah period of about two weeks where i also was basically bidding liquidations and if you go back and look at the chart it would wick like one and a half percent two percent because some guy would get just stopped out and nobody was stepping in on the book and my my logs and hyper liquid it was so funny it was like every trade was like price improved price improved like yeah we were getting the adl on both sides all week and that was pretty eye-opening for sure
Speaker: Was there actually alpha in when your friend told you like the the bombs were falling? Like was that faster than the news? Yes. Yeah. A hundred percent by like about the first 20 to 30 minutes. I had video. i have video from him. Yeah. Because he was living right there. You know, like like he lived in Bahrain. He was like, yeah, this like that was US aircraft that just flew overhead. Like this shit's popping off. Oh, okay.
Speaker: Yeah. It's very bizarre. Like the kind of people that you can meet online. No, I just would have thought like nowadays that news are kind of faster, you know, because like everyone like it it kind of feels like stuff spreads really fast usually. Yeah. Because everyone has a phone and everyone can like record, take pictures or whatever if they live somewhere and just like post it.
Speaker: So I would have thought that it's like pretty much instant now, but it's it's very interesting that there's still, I guess, alpha left in. Yeah, sometimes. having contracts on the ground. Yeah. Yeah. Well, it was funny because that was when, i was it Citrini or whatever, they were posting at the Analyst in the straight, like that started not that long after that. And like, well, I got my guy on Telegram, man. Your own Citrini Analyst number for you.
Speaker: yeah Yeah, no, it's it's the people that you know through the worst contrarians telegram group. yeah That's where the real alpha is. That's where the real alpha is. you ever do you ever do like any discretionary directional trading? Or you just like looking for for arbitrages and more commentators? Yeah. like this this like rare that Not rare, but like ah but what what is like the correct adjective to describe these type of edges?
Speaker: Um... I don't, short they're very short-lived. It's kind of like an information ARV mostly. I don't do too much discretionary.
Speaker: the What I do now, I guess, what you could class that as discretionary with with getting so into AI and being very ah big on like securing hardware and some of these companies that have lots of hardware doing well.
Speaker: Like I would say that's ah discretionary. But the only the only ah coins that I feel that way strongly about in crypto are Bitcoin and hype. And even then, I i find like Bitcoin to be much more attractive than hype at this stage.
Speaker: And when you compare Bitcoin to hardware, like for me personally, I would rather have the hardware because of the utility function of like, I know if I have... more computers I can do things with those in the meantime and it's actually very tough to decide like well should money go into Bitcoin here or should money go into more compute who but that is discretionary I don't like that tape the take. You're like disrupting our our narrative here of the money flowing back into into crypto. We need some opium left. like anything to
Speaker: I think Bitcoin will do well in the long term though. That that is ah pretty much a given. And it's all time preference, too. That's that's another thing. i've I've met people that have done so many different things like meme coins, NFTs. The guy that I interned for when I first got into crypto, he was very prolific new pair sniper on Ethereum. He was doing lots of that was very big memes on Ethereum.
Speaker: And I've seen people make money in so many different ways. And I think that there's always markets that you can go and make money in. And my background was in software, you know, even before I got into crypto. And now that AI is actually getting to the point where you really can do a lot with it, especially locally.
Speaker: I'm leaning into that because i know that I know that I can produce multiples of returns and just do so much in that space. And if you're looking at crypto, it's like, well,
Speaker: You know, I don't see a world where that stuff, like, I don't see a world where AI does good and crypto doesn't do good or vice versa. Like, I don't see a world where AI fizzles out and then crypto goes on a big run. Yeah. Like, all of these, I think markets are going to do very well overall, but I'm...
Speaker: i'm I'm like the worst. I'm like an anti-trader now. I'm like, it's not just the money. Like I have i have other goals in addition to money right now. and i I mean, it does it does make sense if you have like this background and also like ah it it kind of depends on what kind of purse you are, what kind of skills you have. Because if if you can do more with the with the computers, then it obviously makes sense for you to do that. But I guess one of the advantages of Bitcoin is that you don't have to like It's just an asset, you know, you can just hold it. It's not like gold. You don't have to store it anywhere. 100%. You actually have to, like, put work in it, store it, and all of that kind of stuff. So there's, like, different different needs for different use cases.
Speaker: Yeah, I think AI is is going to be very good for the economy, like not just in the, the you know, funny like hand wavy sets where the numbers get bigger, but mean like at the actual amount of value that we're going to be able to create. And this is obviously very good for Bitcoin.
Speaker: Like this this type of environment is is going to be really great. um Do you think there's actually like, ah I think there's actually a very interesting discussion like that, how big are the productivity gains from AI actually, because ah I find it kind of difficult to gauge how much impact it really had on the economy so far, you know, the Fed and Trump and everyone is always like very bullish. Yeah, but I also kind of like the economy kind of depends on it, I feel because like nothing else is really growing. Definitely. So they they they they needed to do well. But I'm always curious, like, how... Like, did did we... Are we actually producing more stuff? Like, is is the output actually bigger than it was before? Because it's very useful. But, like, what what what really changed since AI got, like, that much better?
Speaker: Yeah. Yeah. Well, so there's a couple things about with that. And I actually wrote an article... um a little while ago, back when SpaceX did the IPO, and it was a very hypothetical thing. like I wasn't taking a side about where the economy would go. was saying, like well, you know whether AI does well or not, we can't really predict what's going to happen with the economy.
Speaker: and I said, like for example, they could screw up. credit but Creditors could underwrite stuff wrong, even if AI is really successful. That's true. I'm more bullish than I used to be, and the reason the reason why is because the pace at which it's improving has actually accelerated. And I think a lot of people are not in a good position to to judge the the progress. And like one reason is a lot of people only use the free version, like just straight up, or they used it a year ago or two years ago.
Speaker: And it's like unrecognizable. you know ah When I first got to college, GPT 3.5 came out. like I remember using the first version. And back when that happened, it was the research preview.
Speaker: And um the context window was so small that when you typed a message, it instantly, you got your response immediately. like There was no thinking. It just came back, you know, and you couldn't really do programming with it.
Speaker: I remember like had stupid Java courses at school and like it couldn't write the Java and it was just insane. And now this has since improved, but a lot of people are not using, even the ones who are paying for AI, this is another example, they're using it inside cloud code or codecs.
Speaker: They're not using it inside a good harness. And this is another trap because most harnesses on Twitter are just bad. Like there's a lot of slop. There's a lot of people who are like wasting time building things that are, have bad patterns or just like aren't particularly useful. But it's true that the, the frontier labs just don't have like,
Speaker: particularly great harnesses and they don't have a lot of incentive to focus on that because they're training the models and they're buying hardware like they have a totally different set of things to focus on but then yeah even if you find somebody who knows how to use even just in like codex or cloud code they know how to use the tooling they understand how models work well most of them are only doing one thing at a time like they don't actually know how to parallelize work they don't know how to use like lots of compute And they're dancing around at the top layer of abstraction. They're focused on making like dashboards and like simple apps to sell to people. They're not doing low level things like kernel optimization and just making things like truly better.
Speaker: And now we're starting to see that the models are good enough that they have cross-discipline um intelligence. Like, they're getting truly good at math and logic. And the rate that the models are improving is increasing. But the the problem is people aren't using them in workloads that demand that increase. And it's also very difficult, I think, to conceptualize...
Speaker: um very large numbers, just like scale. Like if you have something that performs at 95% versus 96%, the thing that performs at 96% on like whatever this benchmark or thing is, will be much better for a long running project where you have it completing turn after turn, because it means that it can run that much longer without producing a mistake. And when you start to scale these things up, it's like one person can use pretty much unlimited compute.
Speaker: And so it starts to become a question of, do you have systems design brain? Do you understand like how computers work at a fundamental level? And your, do you have enough like creativity, neuroplasticity, whatever to just like ask the right questions and then be willing to run down those roads of like, there's all kinds of things that I bet that you could figure out right now. with a couple hundred thousand dollars in compute where you're hosting your own models like a lot of these chinese models are very good at this point and they're totally open source but if you go talk to an investor they'll say well what's the return on investment what am i going to get out of this how are you going to monetize that and they want a one-year plan a two-year plan a three-year plan a five-year plan And the things are so non-stationary that it's it's difficult to predict that.
Speaker: But this also invites um this also invites grifters in the same way that crypto does, where because it's ah kind of like nebulous now, they say, oh, well, we can do anything. And it's like, well, I agree, but they're full of shit. You know what mean? Yeah. Yeah.
Speaker: I think it's also part of like what big the the marketing around this. The problem is that the end of the world narrative is like part of the marketing of like the the Frontier Labs. that is Yeah, it's horrible. We have this and that and whatever, and they're ah pretending like it's a serious thing. and maybe Maybe it is, but it's also mostly just to like sell more of their stuff.
Speaker: But I think yeah it's very interesting what you talked about, because like when I talk to irl friends, then... They use AI, but they use the free version. They don't really know like what cloud code and codecs and whatever is.
Speaker: They've never really used that. And on the internet, it always seems like everyone is like doing tens of thousands of different things. But like normal people, I think, don't, like as you said, don't really understand what AI actually is or what it does or what you can do with it.
Speaker: And yes even I, like i'm i'm I'm not really like an engineer whatever. I'm i'm using OMP since you shielded it to me. And it's nice, but I'm probably also not using it to like its full... full excited what what what sort of stuff do you think like What sort of stuff do you do with like parallelization and like all of this compute?
Speaker: what What do you think like people can work on with that? or what What progress can it make that will like actually really affect the economy? Well, another thing that I think is going to happen just generally is because the models, like fundamentally models are just weights. It's just like a graph that you load into memory, you know. um The models themselves are getting released by China and open source stuff does come out of the U.S. to an extent. Like more or less, the model itself is not the valuable bit.
Speaker: It's the hardware that you run it on. And I think a lot of the companies that are going to do the best are like Google, you know, NVIDIA. Obviously, they're making GPUs. They're making ah stuff for the data centers. But Google is a perfect example. You know, there's that guy on Twitter. um His name is For his handle, it's like, you know, user RoyGBiv, something along those lines.
Speaker: He talked about before Google has multiple times more compute that's relevant for AI than the entire country of China. Like, that that type of stuff is not being priced in, broadly speaking, you know? And they're catching flack from finance guys for investing more into the hardware stack. It's like, yeah they are going to do so well over the coming years. Like, they could completely mismanage their company, and they have so much compute that this will just become...
Speaker: more in demand. But a lot of the advancement, the progress that we're going to see, I think will happen in the private sector. And this is something that I believe really strongly because the issue is you can do so much with computers. The question becomes, well, what do you do? And like you said, like, how do you parallelize what tasks you focus on? These very large companies ah Like even the labs, the labs are quote unquote like lean, but they still employ thousands of people. That's thousands of healthcare care plans. That's bureaucracy.
Speaker: They have cap tables, boards, like they have so much stuff they have to go through to approve things. They have the whole safety narrative to worry about. If you're running a company with like less than five people or you just have bought stuff and you're self hosting, which almost nobody knows how to do in the first place, like actually deploy local models on their own hardware.
Speaker: you can just research something if you want to. like If you want to go and research biomedical stuff because like you have a stack that you're running and you're interested in new compounds, like there is nothing that stops you from doing that.
Speaker: And as our software gets better, like we've spent 40 to 50 years building garbage software on top of hardware, and we've kept making the hardware better and better. And I think what we're going to find is if you know what you're doing, you can peel back all of that garbage, and that's going to make your compute retroactively more capable.
Speaker: And so, like small teams, I think they have opportunity to break into all kinds of existing and new industries because there's no way, like OpenAI cannot possibly have verticals in every part of the economy. Like they already make almost unlimited money either by doing training or selling inference. And if they need to sell more inference, they dial down the training and they serve more of the compute. And if they need a better model, they dial back that and they might quantize the model you get or they slow down how many tokens, you know, it gets served to you at and they go train more and they just spend more money. Like they have no shortage. Meta is another one. like Like Mark Zuckerberg has just unbelievable amounts of cash flow. Like it's hard to get your head around how much money these companies have. And so why would they ever go down and do this like some obscure thing? Like like if you wanted to get into, for example, of robotics right now, I think Small companies could do all kinds of crazy stuff in robotics because yeah as our software gets better too, your hardware requirements go down.
Speaker: And if you can run intelligence or you can run very efficient software on simple computers that opens up the door to for you to work with smaller suppliers, smaller computers, maybe make your own. like There's just lots of little opportunity that is going to turn into big opportunity, I think. And I'm trying to figure out what the implications of that are because if you're a private company that's not publicly listed, how is I wonder where that value is going to go.
Speaker: Because it's not like like, obviously you can make money by buying Google, you know Google calls or just like being long Google in general. But if some small robotics company starts in the United States and there's like five people that work there and then they go and make like...
Speaker: $200 million, dollars you're not going to have direct access to that. Like you might benefit from it, you know, a little bit, but you can't just go on in exchange and get exposure to a company like that.
Speaker: That's actually very interesting to think about how how this kind of work for like markets and capital and stuff. what do what do you What do you think about OpenAI and Anthropic then? Do they still have like a ah mode when there's like open source models that are better and cheaper? Is that actually like the case?
Speaker: Do you think that the IPOs will flop? or So i i think that they'll do good in the long term if they have hardware. you know And like OpenAI i for sure is making moves with the hardware. Anthropic, I'm not as sure. I'd like to see them not do so well. They annoy me. like Some of the things that they've done and just the the way that they approach like the safety stuff too as well, like I think they're a little bit worse with that. But they also have people like Lutnik involved.
Speaker: So um I don't want to say on their record, Anthropic is going to flop, man, because they might not. like Crazier things have happened, you know. And there's always the option, too, of acquisition. Like somebody could acquire them.
Speaker: They could get bought out, you know. Like it is a perfect example. A lot of Anthropic stuff runs on on Elon shit, you know what i mean? like Like they're renting and and Elon's got a lot of GPUs. Like what if Elon buys them?
Speaker: He's done something like that before. He's got a lot of beef with Sam Altman. He bought Twitter. He's running SpaceX. He's building TerraFab. Like, he's got, he got away with murder with SpaceX, right? This is another guy who has, like, Mickey Mouse amounts of money. He could just buy Anthropic if he wanted to once they're public.
Speaker: So I don't really know what will happen, but like I do know that the companies that have more hardware are going to be the ones to win because at scale, they can they can make more. They can afford to play with the margins however they want. And the models themselves, like most people already have more intelligence than they know what to do with.
Speaker: They see like GPT-6, like Astra versus Sol versus Luna, and they like like go, I don't see a difference. It's been the same for a year. I still don't know how to use it. And some of it's like user error. Some of it's just maybe they don't have a job that requires the extra intelligence. But those kinds of people could use models that are local now and probably be okay.
Speaker: And so the value is really going to accrue to the hardware that you run it on.
Speaker: i I have a ah maybe half-thrown side question. okay if if this is if If this is all true, what why does the Twitter website not get better? if if we have so much compute and AI is so smart, what why can't they make Twitter better? Well, so, you know, here's and here's another thing, because I was talking a little bit about the public-private company thing. This is another real truth. The bigger the company is and the more established it is, I think the harder it will be for them to update. Like, we are probably on a five to ten year timeline.
Speaker: where the whole economy gets like drug into this new tech. But in the meantime, if you're smaller, the the smaller and newer and leaner your organization is, the more mobile you are. Like the more employees you have, the more the mere bureaucracy is in the way, the harder it is. They're also starting with like established garbage. And a lot of these companies that have a lot of money that don't know what to do, they need a full restructure.
Speaker: um I think what's going to happen is they'll start by cutting workers. like They'll lay off some people, and then you know the systems engineers and stuff like that who are very capable, they'll probably be paid even more money. like there's already You can already go and get paid over half a million dollars a year at Anthropic in some of these roles. like There's already crazy salaries, but I think that sort of trend will continue.
Speaker: um And eventually, you know, they'll get there. But anybody who's worked around like boomers or in established companies and stuff like they'll tell you you, know, these things don't make sense. They waste money. It forever to get done. And so there's going to be a lot of that.
Speaker: And that's why I think the opportunity is so tremendous for individual people. It's kind of like how the Internet was generally a big opportunity and crypto has been a big opportunity for people. And even in crypto, I mean, I think like AI is is a such a tremendous tool. Like a lot of these systems that wouldn't make sense because you go, ah well, the documentation changes all the time or you know there's I'm trading on too many exchanges or or whatever. Well, now if you really know what you're doing, you can rapidly build infrastructure to do what you need to do.
Speaker: So there's there's lots of areas that AI can be applied um But for for most people, I think it's it's sort of like a mentality issue. They don't know how to engage with the tooling properly.
Speaker: And it's it's like a very different type of engineering or thinking from what you do normally. And like when I work personally, you know, I have lots of projects going at once. And I have to be... You have to be able to...
Speaker: handle something and then, you know, change gears completely and go look at something else and not have your original project like nagging in your head, occupying brain space. Because if you set something off to go work for like an hour, two hours, three hours or a day or a week or something while it's working, you've freed your mind up to go do something else. Like you're buying time with the computers. And so being able to switch gears like that so quickly and often is, is really tough. Like it, it reminds me of the interviews where you talk to like guys who used to trade memes, like a lot and their attention is just fried. And then like they, the way that they, you know, process information and everything is like super rapid. It reminds me a lot of that.
Speaker: Um, But it's a little it's a little different too because you have to actually be doing things that are useful. Like so many people are just building dashboards that don't really accomplish anything. They're not scaling a business or improving what they can do on local hardware or or learning.
Speaker: And like now it pays to be a a generalist instead of being overly specialized. And you have to be able to think about how to build good systems out and how to automate things. And those type of like autistic people who play Factorio or those sorts of games where they're constantly but like that's the exact mentality that you need to do this sort of thing. And it pays dividends because you. There's so many jobs that really like probably shouldn't have been jobs to begin with.
Speaker: But now with AI, it's like, forget about it. You can genuinely automate so many things. And it's very jarring for me to get onto Twitter. And ah like I don't know how to explain it to people because there's there's grifters that are on Twitter. and i don't want to sound like a grifter. But like I talk to people and I look at what I see on my For You page and people are saying, like you can't do this or that with AI or this model is bad. I'm like, that's not true. like I know this isn't true because I've done. like what What you're saying can't be done. I've done it. And i I think some of it is an intelligence thing, but some of it is also like a mentality thing. And it just it depends on what you're talking about. But overall...
Speaker: it's It's a very unique situation. And I think you're going to find as well that just like how a lot of the kids who grew up playing some of the newer consoles are just unbelievably good at video games because they had exposure from a young age. I think people who don't have a lot of priors about how long things should take or how stuff should work, I think they're set up really well to do things with AI. Like I'll do projects where...
Speaker: ah There's no documentation of it being done online. And you talk to the AI and the AI says, either it can't be done, like this is extremely optimized, or it'll say like, it'll take a year. And I'm like, okay, well, we're going to do it. And then it's done tomorrow. Like, like something's not adding up here. Yeah.
Speaker: I mean, I guess it's kind of like any new technology a bit. Like, yes when it first comes out, people don't really know what to do with it, how to utilize it properly, like with the internet or even with like fire or or some stuff like that. Like, you when when humans first got control over fire, they they did a lot less with it than is actually possible. That's really learning continuously throughout human history. So I think AI is kind of like,
Speaker: it It makes sense. you know It's like so overwhelming in a way as well because it can do like so much, the potential so great. yes But you really need to like just play around, experiment and stuff. Do you have maybe, ah like to to come back to training a bit, like any specific examples or advice for people, like what they can do with it for crypto or markets in general, like use it for research or like yeah automate stuff?
Speaker: yeah i'm so i ah first of all very straightforwardly um data you can parse so much data now because of this um you're able to do sentiment analysis on social media that's like very trivial on chain stuff you can stand up things in you know minutes hours like like very complex systems you can parse large amounts of data um And reading documentation, this is another thing. Like so much of what I did in the last year or two, it was me at my computer, scrolling Twitter, talking to people in Telegram and then whatnot, and then finding documentation and reading it. well
Speaker: If I had scrapers back then, like if I could stand up scrapers easily, that could just read the documentation and monitor the documentation for for changes. You know, like maybe you've already maybe you're trading like commodities on Hyperliquid and you just point something at the documentation and at the social media accounts and you've got a small model just kind of parsing. that information, looking for changes in the page or tweets and searching for relevant information just in case, you know, just in case you missed something where they've made a change or they announced something new.
Speaker: And that type of stuff is super accessible. And a lot of people I've heard in the past have said, like, you know, you have to make the decision to Utrade on the big centralized exchanges, like maybe you trade on Binance or do you trade on a spattering of smaller decentralized exchanges? Well, now you can build systems so easily, it's kind of not really as big of a decision to make to go trade in a new venue. Like, of course, there is still some trade off, but be When you figure out how to ah build things in parallel or or how to build systems that last, like a lot of people will ask for the same thing over and over again instead of getting AI to build a script once that does that thing for you repeatedly. And it's like being able to think in systems, that's so huge because it allows you to behave like a more competent player in the space. Like standing up simple bots now, this takes no time at all.
Speaker: you know the The alpha side of it is a little bit more complicated and you have to develop your own strategies maybe or you have to steer a model if you're doing research and you're getting it to put together data packages for you to review.
Speaker: But overall, it allows you to get so much work done. It's like you know imagine if you had, like not to use like any type of marketing or whatever, but it's like if you had interns that you could use to do work for you, except that they work faster and that they can arguably be smarter because you can develop a script and then just offload it. It's not like having to get a person to do it manually. And you can do this super parallelized.
Speaker: And so you're able to compress like hundreds of hours of research and coding into a very short period if you want to. And so I think there's a lot of edge there in going to look at some of these chains. And quite frankly, we're seeing this with DeFi as well. All these DeFi attacks,
Speaker: This is just somebody, know, these are just people who are using AI, who are running like Chinese models locally, stuff like that for privacy reasons. And they're they are they're exploiting smart contracts.
Speaker: Like that's ah that's a perfect example. there's There's chains that have been dead for years that are that people are just draining and walking away with 10 grand, 50 grand, 100 grand. Like it's free money. it's been It's been sitting there on chain for years.
Speaker: And it's because it's just, there's so much data out there. It's never been practical to go through and audit it. but now you can't. and i'm not I'm not advocating for you know like hacking protocols or anything, but um you know there's lots of stuff that's out there that you can do.
Speaker: Do you think we'll we'll all be docs eventually on-chain? With AI being probably. Or there's enough compute and stuff that you can just like check everything. Because I think most people have like treated it as kind of like... I mean, on-chain is not anonymous, but like yeah you can like obfuscate yourself a little bit. and like People didn't really bother to look into it because it's just like too much like it's too much. It used to be before AI at least, like way too much effort to like really...
Speaker: You couldn't even like look at your own stuff if you used different chains or whatever for text reasons and all of that shit. But I guess now with AI, it's probably quite possible to have like a lot more transparency in that regard. Yeah, I think...
Speaker: I used to be extremely privacy pilled, like in general, you know, this was something that was important to me and AI has changed this in multiple ways. Like one, it's so easy to get information and we've been living like post nine 11, like there's been so much, data know collection of data, but, uh,
Speaker: One thing that's nice is the companies don't care as much about your data now because they've already got it all and they're using synthetic data for everything. So it's kind of like a paradox that people care less about individual consumer data now. That's one half of it. The the other thing is I've transitioned more to like security through obscurity.
Speaker: like you're You're never going to be completely anonymous to like a nation state actor or something. Yeah. But you know there's there's stuff that you can do that won't draw as much attention to yourself necessarily. and like the The whole thing is a spectrum, and it always has been. But I think this is more true now than it was before. And with everything you do, with any information you give up or things you say, you know its if it's on the internet, like the public internet, it's going to be out there forever. If it's on chain, this data is never going away. And the things that you say, and like it's just... I think being measured...
Speaker: with what you say and do and what you choose to share is is important. But yeah, at the end of the day, I mean, Like there's Monero, I guess, you know, you can use Monero.
Speaker: and But even then, it's just we never know what's going to happen, right? Like what if some of us are pretty young guys. it's like Exactly. Like what if what if they break it open and not permanently? Like i'm not saying, oh, there will never be security, but you you don't, there may be a day where some of this information that we think is locked behind 256-bit encryption becomes public for even people.
Speaker: you know, five minutes, 10 minutes. Like it doesn't take much. And once it's recorded, it's recorded.
Speaker: Yeah. Well, I guess we we just have to wait and figure But I wanted to ask, like do you have any... like Because you were mentioning to like think more in systems, and I think it's quite it's not not so easy to like get to a point where you can do that like very well or properly. I guess that's also that a lot of high-paying jobs, or like those are like very high-paying jobs because it's not an easy task to do. Because I think like many of these things like markets or like building stuff is already kind of like difficult to approach. Like you yout you need so much information to really know, kind of like to orient yourself on what you're doing.
Speaker: And then like properly thinking to like create a system around that is ah quite difficult, I think. Like, did do you like any advice on how to get more into those, into that mindset or that type of thinking?
Speaker: Yeah, it is tough. i like like My background, i when I was young, I have learned how to use Linux. and and not Not in like the meme-y, ricey way, but like my dad had old college textbooks, and I would read them, and it was like how to use the the terminal, and set, and awk, and basic commands, and things. And so I learned how to do a little bit of programming, shell scripting. And then I really didn't do much until almost when I went to college. But I had this i had this fundamental base of understanding how computers work.
Speaker: And then I've taken, over the years, I've done like you know some logic design things, electrical, that sort of deal. And having that engineering background ah was was really the key to being able to understand that. I think, you know, some people are naturally just going to be better at it than others, but ah it does pay to understand everything.
Speaker: in a general framework, ah how do computers work? You know, how does logic work? And then when you're building systems, I mean, like really like one of the reasons that the models have gotten so good at everything is just, it's just math, you know, like it's, so we're we're basically just backing into logic and fundamental logic. You can do whatever you want with it almost.
Speaker: Um, And so it's just thinking in those terms. Like I brought up Factorio earlier, stuff like that. That's literally what it is. You have some kind of thing that you're optimizing for. Maybe you're trying to reduce a number or increase a number, ah you know, and and then you you start working down that path of how do I do this?
Speaker: in the smartest way and so if you're building a trading system right you have to think about your core strategy and then you have to think about your integrations and how you hook it up to the exchange and then there's everything in between that like you have to handle if the exchange sends something bad or if it goes down or if like there's some kind of glitch like how does your system deal with that you know that's that's one thing and then aggregating your data maybe you have multiple ah exchanges that you're trading on you have like a portfolio and you have global risk you got to pull everything together and then you know maybe you think about redundancy and so i've got one computer while i can have and another computer that's mirroring a lot of this just in case so it's it's stuff like that um but it's it really it really is just an an engineering mindset fundamentally and if you're good at it in one domain
Speaker: you can kind of just apply it to others. And this is even more true now than it was before because with AI, you know as the models are getting smarter, you don't really want to lay out necessarily how to implement everything down to a T. like the The inference is there for a reason. And so because you can lean on the models, having general broad knowledge enables you to think of better things to do with them. And from a trading background, like the easiest thing that you can do, not even just dashboards, but you start thinking like, what is everything that I do in my job? Like, how can I break up all the markets that I'm engaging in? I have a research component, I have a trading component. And and then you start saying, okay, can I build systems out of these?
Speaker: And like, if you don't know where to start, the easiest thing to do is trying to automate everything that you already do. Because you probably can. And in the past, this was difficult because software took time and money to create.
Speaker: But now, if you just have money even, you know you can rent compute, you can buy subscriptions. They're still very cheap. like You can do a lot and you can just directly convert your money into work output.
Speaker: And that wasn't a thing before. So it opens all kinds of doors because we're still... in many ways we're still really early to this whole like ai thing and like i said earlier five to ten years like this is going to take a while for everybody to really kind of understand you know where this is headed and what we can actually do with this stuff But as an individual, like if you're a trader and you're managing your own capital and you're doing your own thing, I mean, you don't have to tell anybody what you're doing.
Speaker: And I think that's another issue with the AI stuff is the people who are perhaps utilizing AI the best are going to talk about what they're doing the least. Yeah. Yeah. Yeah. It's like alpha leak, you know?
Speaker: Yeah. They have no incentive. And then if you do, they say, you're an idiot. Mm-hmm. here I have a bit of a counterpoint though to to like what you just said, but or maybe not like a counterpoint, but I've been thinking a bit about... um and like i've I've read some like an interview with Koala recently and and also stuff that Robot James writes all the time but whatever. um yeah I think AI is also really useful like if you really just like want to get started with stuff. Because I've like done a lot of experimentation this year on like automating shit.
Speaker: But um I think it it, first of all, it helps to like properly understand knowledge that other people give you because sometimes there's just like like other people write articles whatever and then there's like some stuff missing that I don't really get. It can help you to understand that better and research that. And also like immediately kind of turn that into strategies and back test that. I think it's very useful for that.
Speaker: And also... um Because they they always go on about like, don't try to automate stuff first, but like just try to like execute stuff manually at first. Because I mean, I think now with AI automation is why like really good, but also it's like just helps you if you kind of understand the intricacies of what you're exactly doing.
Speaker: Especially if you're like 100% more of a beginner, you know, you need to like you know a bit about the microstructure and whatever. And I'm also like shilling the the terminal for manual trading a little Yeah, yeah, yeah. yeah I think it really helps to understand because I like I i did.
Speaker: I have like some automated stuff, but it's like 100% vibe coded, you know, and I have like a little bit of a coding background and I understand like its strategy part of whatever, but I've never looked at the code. I don't actually know how the execution like properly works and stuff like that. And I think Maybe it ties a bit into what you said before, like really understanding the the details of what you're doing, like with the computer and logic stuff and whatever also really helps you in just everything pretty much. It's very hard to give advice because i'm i am who I am because of everything that I've done. And yeah I did some manual trading first. And I think that's a super valuable experience because you learn a lot. Um, and all this stuff is always changing to, you know, whatever, what everybody has access to and what they can get out of the tools as an individual might be different from me.
Speaker: Um, but yeah, no, I totally agree. And I remember I had a point that I was going to make there, um, with the automated trading. Oh, as far as reading the code. So like I've done some incredibly large software over the last six months, I run,
Speaker: systems out of my house that are essentially like entire SaaS companies, but just for me, for like specific use cases that I have yeah and things. And I really haven't been reading the individual code at all for over six months now. But what I do have is...
Speaker: ah extensive, like all of my repositories are HTML documented and they're very well documented and it's like hyperlinked and I can go through and read it. So I have an understanding if I need to of the architectural stuff and obviously for trading systems, like, you know, you need to know what the main architecture is. But in terms of the actual code, like code monkeying stuff,
Speaker: If I have a problem or if I have a feature that I want to add or I'm rebasing something or whatever, I'm i'm always moving through a model. Like I'm really not using text editors anymore. Like my my entire workflow when I'm managing things is I'm just in a terminal that has a multiplexer so I can have lots of different tabs and I can go through and I can see. you know, different sessions and I'm running OMP in those and I'm, you know, talking to a model that can go manage other things. And all the computers that are in the lab at my house are clustered together. So it's it's just a giant pool of compute that my AI can use. It's like yeah all the GPUs in my house, all the CPUs, all the memory, like this is just all their storage. And it's it's just very different, you know, from what like a typical developer workflow has been.
Speaker: And I talk to lots of people from different backgrounds. Like, obviously, this is a crypto account. I have lots of friends in crypto. And it's very strange to go back and look at DMs from like a week ago or two weeks ago. And I look at the advice and I'm like...
Speaker: This is horrible. Like, it's not true anymore. You know, like everything we talked about is not the case. It reminds me a lot of crypto. Like when you tell people like, this is the meta, like you need to be trading on this chain, these memes right now. And two days later, it's dead. Like you get eaten alive.
Speaker: i but what One more random thing that I want to what i briefly ask about is, I guess, I don't know if you experience this because I don't know how how like your personal workflow looks like specifically, but um like I sometimes run stuff in parallel or whatever, but most of the time I just like have one one thing running at a time, more or less, or like im I'm doing like one task.
Speaker: And um I don't know how to like properly explain but I think there's maybe like a little bit of a Slopification of work. Almost. If that makes sense. Do you understand what I mean? Because it's a bit like. like Doing stuff with AI. Is a bit like scrolling TikTok.
Speaker: yes Because it's like. You put in your stuff. And then you kind of like wait. and you're you're just like mindlessly like engaging in stuff a little bit. Because it's just like you you're not like it's not like crafty. you know you're not like like i don't know You're not a smith and doing stuff with your hands or whatever. but You're just like walking input input, input, input. I slightly disagree. So what I've been saying lately...
Speaker: is if you if you're actively engaged in one session, you know especially, i mean, I guess this is true all the time, but especially if you're using a fast model that does like hundreds of tokens per second and like the output's coming out faster than you can read it, you know, um I feel like a wizard. I feel like a modern day wizard because I don't type anymore. i just use the stupid like whisper flow voice thing. Because you know the way you think, it's just like when you write with your hand versus when you type on the keyboard. like When you're talking out loud, it's it's a different experience and the way you think is different. But it's almost like you're casting modern day spells. And like I say that genuinely because you're... you're
Speaker: You have people talk about the gnomes problem and they say like, well, you have step one and then step two is like the question marks and step three, you make money or you get your thing. And they say, well, what's the middle? That's literally the ideal workflow for AI because what you're doing is you're talking. You're like, I want this and blah, blah, blah. And here's these things. And if if you know...
Speaker: what you're talking about. like It's like a combination of being a good orator and having broad general knowledge. like Even when we're talking ah in English or in different languages, because different languages have different relationships between words, they mean different things. like There's lots of semantic links between stuff.
Speaker: And a language model is literally just that. It's weights that are tuned on this stuff and you put things in and it just generates an output. So if you have the right idea of what you're looking for and you have lots of experience, like I've spent...
Speaker: eight to 12 hours a day for over a year using these tools, and I've had experience with them since they first became mainstream, like in 2022, you develop an understanding of how to use them effectively.
Speaker: And that enables you to just like, I can i can prompt things and and get what I want out of them. And that was a huge disconnect that I noticed when I went on Twitter when ChatGPT 5.4 was when I noticed a really big shift in capability. But even looking at like stuff that we've had lately, 5.5, 5.6, like Sol as an example, I've had multiple accounts in OMP on ChatGPT, and i could give it a prompt that's like,
Speaker: just odyssey of a prompt. And it's like a few paragraphs. And this thing runs for days, like like literal days. And then it comes out and it's exactly what I asked for. Like I'm like, I wanted this whole code base rewritten into Rust just because. Like I just want to benchmark it and see if it's better or not. And I gave it and it was like a whole trading system that I had and it went through and it was done.
Speaker: And that kind of value, it's really difficult to put a price tag on that because first of all, it only costs me $200 a month per subscription. Resets Weekly, and I had a couple of those. But this whole system, like, yes, it took days, but it took me, you know, an hour thinking to think like, okay, what do I want? And then I go and do the prompt, and I sit there, and I i talk for five minutes, and then it's completely out of my mind until I get the notification on my phone that the server is done, and I open it up. And this is such a bizarre way to work. Like I think the five day work week standard stuff is totally done. I'm getting work done at like, you know, dinners, like out with friends and stuff like that. you can just check things on your phone. It's very, very different than like old school, like sit down at the desk, eight hours engineering, like figure it out.
Speaker: Yeah. That's kind of what I was getting at. Like the the way that you work is like very, i don't know, just, just different. Like the five, five day work week and stuff like doesn't really make, make sense anymore because it can just like,
Speaker: and I don't know how to properly like verbalize this, but it's just a yeah very different way of of engaging with with your work. if you like If like some of the thinking is outsourced because that has never really happened before, like not in this capacity.
Speaker: Yeah, I think we'll see some real changes over time. It's it's going to be both slower and faster than people think. Like there will be old companies and old ways of doing things that will stick around for what feels like forever.
Speaker: Like five years later, people will still be doing weird nine to fives and like the world's not going end. But at the same time, in a year or in six months, like you're just going start seeing things. You're going to start seeing things that that you wouldn't think like could have been done before. And that's super obvious to me from using it because I'm just one person.
Speaker: I don't know everything, you know, and I can only focus on so many things. I'm only going to make so much money. And there's so many people out there in in the world. And it just it doesn't take that many, you know.
Speaker: Like if a thousand people figure this out and go start companies, like imagine the the kind of stuff they're going to be able to do. And you see this with trading too. You see like people on crypto Twitter who are around for like six years and you've never heard of them. And they've made like a whole bunch of money. And they've just somehow been in like projects you've never heard of, you've never seen. And they've been on Twitter the whole time. It's not like they're in the middle of nowhere.
Speaker: But you just never even run into them. I think the pace of stuff is like very interesting because to me, like like i um I'm starting to get to the point where with how I think about this stuff where I think what everything that you say makes a lot of sense.
Speaker: And I see like the usefulness in the tools that they're really like improving and stuff compared to like six months ago, 12 months ago. yes And I find that really cool because the same time, like, at the beginning of the year and, like, last year, I was honestly kind of disappointed with how everything was because they made the expectations, like, so insanely blown out of, like, you yeah yeah when when Cloud Code first came out, like, yo, this will change, like, society completely within the next year and, like, the permanent underclass everything is over blah blah blah obviously there's a bit marketing and dramatization people don't really know how much is actually going change but like when you hear that because we're like online every we twitter and then you hear that all day it's kind of like oh shit is this actually going to happen twitter is such a bubble
Speaker: i mean But it's it's really it's really cool to see like the that there is actually like progress. you know there is yeah it hasn't really like it It hasn't stalled, there is like stuff going on. It's maybe not as extremely fast as you expected, but still like stuff is happening, models are getting smarter, there's more stuff that you can do, there's just more and more potential, more opportunities. The world hasn't blown up yet, like nothing has like been destroyed. We don't really have like yes machines that are alive either, but it's still a very useful tool.
Speaker: It's kind of like the best outcome almost. Yeah. And when you look to, if you pay very, very close attention, there are major optimizations that are happening. It's just like a lot of people who aren't in software, like the Linux kernels getting speed ups and like some of the the numbers on multiples of improvement and people are developing, like when I develop applications now, I just build native apps for everything. Like if it's on Mac, it's a Mac OS app. If it's Linux, it's Linux. If it's Like ah this was just making the rounds now on Twitter that ah the Super Smash Melee game was fully decompiled. You know, I was doing this months ago. I began doing stuff like that with far smaller models. Like people are saying now, oh, GPT Astra is like finally got smart models. We can do this. If you knew what you were doing and you knew how to do like fine tunes and other things, like it's kind of incredible what you can do.
Speaker: with some of these tiny models just for any task. But that type of stuff is huge because that this now means like we can take applications that were never released the code. You can find out what the code is. You can optimize them. You can just port them over to something else. And these type of changes will start rolling out. But when you go to the largest companies, like even big companies like Apple or Google or Meta, They have so much, you know, just weight, like dead weight and then things that are in the way. It's going to take a lot of time, like like Twitter, you know. How many people are really in charge of decisions? And I remember a lot of people criticized Elon for the firings when he first got Twitter and he changed it to X and they laid off a lot of people. But I think i think he had the right idea.
Speaker: generally speaking. i mean I mean, the website is still kind of like the same. like It's not like it was that good back in the day. They just right kind of keep it going with like one-fourth of the people, so it's probably not that bad. It would be cool if they also improved it, but maybe there's too much for now. Elon, if you can hear us, please fix Twitter.
Speaker: make make the At least fix like the search or like whatever. it's like kind of ah But like I think it's also kind of part of... like So scrapping is like not done as easily, but it's like very annoying from a user perspective. He's got like bigger fish to fry now. like yeah he's He's worried about like building like drones and robots and Starlink. He's like doing huge stuff. And he already bought Twitter, so he's like, whatever. like I can get on there and tweet, and 300,000 people will like my posts. That's true Yeah, I think this is a good point to wrap up. This was like a very different episode from usual, but I found it very interesting. I hope our listeners did as well.
Speaker: is Is there anything you you want to mention at the end? and mean um No, not particularly. I'm really glad that you had me on and i I enjoy these podcasts that you do. I find that everybody has a unique ah perspective to bring. like I actually take the time to listen to these.
Speaker: Thank you. That means a lot to me. I appreciate it. Thank you very much for it for coming on.
