[00:00:00] Andrew: On both
[00:00:01] Nathan: Twitch and YouTube. Yeah.
[00:00:02] Nathan: Cool. How's it going?
[00:00:04] Andrew: I'm [00:00:05] good. You're on location
[00:00:06] Nathan: Yes, I'm in a different spot. I'm moving around. [00:00:10] So I'm, I'm in San Francisco, the Silicon Valley area.
[00:00:14] Andrew: [00:00:15] Nice.
[00:00:15] Andrew: And
[00:00:16] Nathan: I believe my qual- my video quality might be low too, so we'll,
[00:00:19] Andrew: [00:00:20] No, it literally just started melting the second we started. So you sound [00:00:25] good on my end.
[00:00:25] Nathan: I, I'm without internet.
[00:00:27] Andrew: Yeah, we'll keep it going. So hopefully if you're watching [00:00:30] live, Nathan may come in and out of existence, but the audio's been really good. If you're listening, you probably won't [00:00:35] notice, but it might be worth watching the live stream a little bit to, to see
[00:00:38] Nathan: I was gonna say, honestly, it's, [00:00:40] it's no different than if I were in Port Rico. I could always just at any minute just drop off.
[00:00:44] Andrew: Well, and it's funny 'cause [00:00:45] we, we kinda joke about it, but it, it you do kinda melt sometimes towards the [00:00:50] end. The quality visually degrades and,
[00:00:53] Nathan: Yeah, yeah. [00:00:55] There, I, I know that on at least two different occasions where at the end I, I think I [00:01:00] hear you like, "Oh, we lost Nathan." I'm like, "No, I'm sitting right here," but it seems like
[00:01:04] Andrew: Yeah.
[00:01:04] Nathan: [00:01:05] I'm, I'm not there. Yeah, you're in, It's funny too that you're in Silicon Valley and it's like [00:01:10] the internet's worse than being on an island in the Atlantic.
[00:01:12] Nathan: Yeah, I was, I didn't wanna make the joke, [00:01:15] but yeah, it's, it's Google.
[00:01:17] Andrew: Google. Own it. But anyway, [00:01:20] yeah I wish they would've kept that whole enterprise going. They still have the internet Google Fiber [00:01:25] stuff, but they basically
[00:01:26] Nathan: Yeah, but it's, it's I don- you heard, or at least I heard a lot about [00:01:30] it, and then it seemed to kinda drop off, and it feels like it's just one of those things where it's just there [00:01:35] now. It was being pushed for a little bit, and all of a sudden it, it's not. And they [00:01:40] they seem to have a tendency to do that.
[00:01:43] Nathan: I
[00:01:43] Andrew: Yeah, like they killed AI Ultra [00:01:45] for Enterprise and Antigravity. I'm not mad about it at all. No. Google, I've [00:01:50] switched my usage completely away from your products now. So good job, big [00:01:55] brain product management there. Trying to make AI a consumer-only [00:02:00] product is super dumb, but good luck.
[00:02:03] Nathan: Yeah, yeah.
[00:02:04] Nathan: And they would've [00:02:05] had a power user in if you were still using their product,
[00:02:08] Andrew: I was probably the only [00:02:10] evangelist.
[00:02:11] Nathan: You're the only one, you're the only one cranking, cranking the [00:02:15] engines there.
[00:02:16] Andrew: Yeah it's still like Antigravity. It was a, kind of a VS [00:02:20] Studio clone. So you could manage everything from there. The chat window wasn't like the chat [00:02:25] window in VS Code. It was actually integrated into the AI [00:02:30] sphere, so it had full visibility into everything you were doing. And it, and it was [00:02:35] more of just like a window in there.
[00:02:36] Andrew: And then the plan The doc, [00:02:40] document, plan, review, iterate, I love it. I still [00:02:45] can't quite emulate it with Claude Code. I've gotten close. It's kludgy at [00:02:50] best, and it's just really annoying. I have to I have all sorts of markdown file [00:02:55] problems and I don't understand why VS Code can't handle it.
[00:02:59] Andrew: So Google's actually [00:03:00] made an incredible product there, but they try to-- They apparently think that there's a huge marketplace for that [00:03:05] for customers, which doesn't, like B2C doesn't make any sense to me at all. [00:03:10] Sure, people wanna like build an app or something, but folks paying 20 bucks a month is [00:03:15] just not real.
[00:03:16] Andrew: I don't know what to tell you. You can't build a killer app for $20 a [00:03:20] month in AI credits or subscription. It's just not real yet, or maybe [00:03:25] ever. We'll see
[00:03:26] Nathan: I think we were not planning on, on talking about this but
[00:03:29] Andrew: [00:03:30] No,
[00:03:31] Nathan: I, I think, yeah. But it's an interesting topic, right? Because [00:03:35] there were... It was Google, Anthropic, and OpenAI trying to push sort of the frontier [00:03:40] intelligence, and then Google sort of dropped off, right? Grok was also in, in the, in the mix as well.
[00:03:44] Nathan: They seem
[00:03:44] Nathan: to [00:03:45] seen a, a little bit of a revi- they've seen a little bit of a revival, but for the most part, yeah. [00:03:50] But we were talking a bit offline about it, but Google's approach going towards [00:03:55] consumer application, I would say, in some way, it's like a blue ocean where no-no [00:04:00] one's quite taken the, the, the crux of doing so and, and [00:04:05] Google seems to be in the right position, in the right position to do it, right?
[00:04:08] Nathan: OpenAI, who had [00:04:10] tried to break out into that space by creating these so-sort of side quest things to[00:04:15]
[00:04:15] Andrew: Nah, it was the whole time. Nah, it was... I don't know what that was. I think it was, like, a [00:04:20]
[00:04:20] Nathan: I think they were, I think they were, I think they were trying to... Yeah, I think they were trying to, to, to amass an [00:04:25] audience by going to the general consumer, and it just didn't work. But I, I don't know [00:04:30] anything about this. But anyway, c- Google sort of dropped off by,
[00:04:33] Nathan: by trying to to focus on [00:04:35] this sort of unmet market, and maybe they'll crack the nut. I, I don't know
[00:04:39] Andrew: I-- I've said it [00:04:40] before, the, the Google Workspace was like perfectly positioned for AI, and I [00:04:45] think they've done a pretty good job integrating it. Is it useful? I don't know. [00:04:50] I'm not probably the target market. I don't think like taking a piece [00:04:55] of information out of my email and putting it on my calendar is worth like an upcharge.[00:05:00]
[00:05:00] Andrew: And
[00:05:00] Andrew: like,
[00:05:01] Nathan: tokens.
[00:05:02] Andrew: right? This is something Apple's been [00:05:05] failing at for so long too that I just I don't... Like sure, I would love a world where [00:05:10] the AI assistant just kinda knew what I wanted and executed flawlessly. The reality is it's not anywhere [00:05:15] close to being there, and anything important I'm still gonna manage myself 'cause it just doesn't work well enough or [00:05:20] consistently.
[00:05:20] Andrew: But I don't know. But the, the, the programming piece, like [00:05:25] they were really ahead with Gemini 3.1 Pro
[00:05:28] Andrew: And [00:05:30] Flash, I don't know
[00:05:31] Nathan: even Nano Banana, they're still yeah, on the
[00:05:34] Andrew: Yeah,[00:05:35]
[00:05:35] Andrew: that's kind of consumery, but in all your Google Photos. [00:05:40] But, the... Mm, no, I'm not gonna go there. The the coding space [00:05:45] though, like they were really ahead and like some people still swear by Flash 2.5 for [00:05:50] specific tasks it just crushes on.
[00:05:53] Andrew: I didn't think Flash 3 [00:05:55] was all that impressive and f- especially for the cost, like it just, like me as pro. [00:06:00] But 3.5 Flash has been [00:06:05] pretty damn good. Like it's g- seems like it's gotten worse, but compared to Sonnet 5, [00:06:10] like it's much more capable for utility purposes. Like [00:06:15] code is still a little hit or miss, but like Sonic, Sonic's pretty good for coding.
[00:06:18] Andrew: I don't use the other [00:06:20] companies, so I, I can't really talk, talk about it. The but like the Antigravity [00:06:25] piece, I, I, it seems like Google has bet on customer and it or consumer, and it [00:06:30] seems like B2C, and in that paradigm, Antigravity doesn't make any [00:06:35] sense and coding doesn't make any sense 'cause we're just not anywhere close to that being real for most [00:06:40] people.
[00:06:40] Andrew: But like the company that invented transformers and shot, shooting themselves [00:06:45] in the foot, I think they're doing that again with like coding 'cause they were a little bit behind. But I [00:06:50] think they could've really pushed it forward because the last Google I/O, they had a [00:06:55] huge like agentic framework push, and they have some really cool frameworks for [00:07:00] enterprise.
[00:07:00] Andrew: It's still a little too developer-focused for like [00:07:05] non-technical or non-developery people to like really execute on, but that foundation [00:07:10] is clearly there, and I'm sure they'll integrate it into Antigravity and other stuff. So it's like you have a [00:07:15] one-click spin-up app once you like build your app or whatever.
[00:07:18] Andrew: But it's a huge missed opportunity. [00:07:20] Like, the biggest missed opportunity, I think. And [00:07:25] Google is always product is always about trade-offs. Google has been pretty [00:07:30] good about shooting themselves in the foot, but it usually makes sense. Like they're trying to go for [00:07:35] scale or they're trying to go for a play in a market.
[00:07:38] Andrew: But for this, [00:07:40] like leaving the coding market like completely aside and like not having a solution, and like the [00:07:45] solution like they do have is, forget what it's called. It's also super [00:07:50] confusing, so great way to sell your customer and stuff is just make it super confusing. That's a [00:07:55] surefire way to get a mass market adoption.
[00:07:57] Andrew: So I don't know what they're doing and it pisses me off 'cause like [00:08:00] I, I actually really do Antigravity still. I just can't use it 'cause it's, it's not, I'm not [00:08:05] paying token-level pricing when I can get subsidized somewhere else, frankly, and [00:08:10] also aren't good enough to pay for when there's o- other competitors on the, the market.
[00:08:14] Andrew: Sorry, [00:08:15] Google
[00:08:15] Nathan: Yeah. Yeah, yeah. We did talk about Antigravity when it first came out, and I, I do [00:08:20] agree
[00:08:20] Nathan: with you. The-- they, they really nailed the interaction mechanisms, I think, for [00:08:25] certain types of people. I think, I think that was the big miss. I, I-- you can [00:08:30] see that with was it CoWork and I forget what OpenAI's is [00:08:35] called, but it's very similar.
[00:08:36] Nathan: But they, they tried to come up with this nicer UI and [00:08:40] a way to, to, to leverage the, the either Codex or, or C-Cloud code. And [00:08:45] I think Antigravity had them beat on both fronts to market and in terms of usability, it [00:08:50] just never-- they never doubled down on it unfortunately
[00:08:54] Andrew: they were pushing and [00:08:55] like I'm all for B2C and exposing things to consumers. I, [00:09:00] I think that's great. And it's like, but the money is in the enterprise right now/maybe [00:09:05] forever. And as a byproduct, if you want to get users and [00:09:10] expose people, that's great, right? Let the enterprise pay for it, offer a private [00:09:15] tier for cheaper price or whatever.
[00:09:17] Andrew: And I, I-- my assumption [00:09:20] was that the margin for like the workspace team was just getting hit too hard or [00:09:25] obliterated 'cause like I was probably obliterating your margin. Sorry, not [00:09:30] sorry. I was using pretty much 100% of my quota like every, every [00:09:35] month. And like you could have adjusted that.
[00:09:37] Andrew: You could have made another tier. I'm waiting for like the [00:09:40] $500 a month tier from a company or the $1,000 a month tier. It's [00:09:45] coming.
[00:09:45] Nathan: it's probably gonna happen. Yeah,
[00:09:46] Andrew: Yeah.
[00:09:48] Andrew: And we've talked about this before, but if you have to [00:09:50] pay an engineer a couple hundred thousand dollars a year plus benefits, like 1,000 bucks a month [00:09:55] for a coding assistant that's actually really good kind of makes sense in some instances.
[00:09:59] Andrew: However, [00:10:00] it has to be good. It's just not there yet.
[00:10:02] Andrew: And I don't, I don't see it. I've been listening to a lot of AI doomer [00:10:05] this last week, and I just, I don't see the doomerism. Yes, [00:10:10] theoretically it's like science fiction. I'm like, sure, it'd be really cool if we could fa- travel faster than [00:10:15] light.
[00:10:15] Andrew: Physics doesn't allow for that, and maybe it's possible and we just don't [00:10:20] understand it. But I'm like, it's a fun thing to think about, but I'm not gonna live my life [00:10:25] be- in a way that is gonna be impacted by faster than light travel, which is [00:10:30] theoretically not even really possible. Quantum, we've talked about this a little bit.
[00:10:33] Andrew: It gets a little weird. But [00:10:35] like
[00:10:35] Nathan: Yeah
[00:10:35] Andrew: ever see a reality in my lifetime where Andrew is gonna transport to [00:10:40] Mars in less than five months. Prove me wrong.
[00:10:43] Nathan: Not in your, yeah, not in your [00:10:45] lifetime. Yes, yes, yes.
[00:10:46] Andrew: And if we do, great, I'm all for it, but I just-- there's no evidence, [00:10:50] right? There's just none. So yeah, I, I, I don't know. I guess [00:10:55] just a hint on the doomer stuff, it doesn't cease to amaze me, but I [00:11:00] find myself continuously frustrated by the way humans are operating in this [00:11:05] era. We've never had it better in almost every instance, and yet we [00:11:10] find ways to continuously sabotage ourselves. So New York just put out a bill that [00:11:15] pauses all AI data centers for a year because a bunch of [00:11:20] reasons, frankly.
[00:11:21] Andrew: And we've talked about this before, and I've taken a lot of heat for it [00:11:25] on and offline, but it's all made up. It's-- The AI data center problem [00:11:30] isn't a problem because electrical generation and distribution is cooked. [00:11:35] It's, it's booked out for five or 10 years. They could build all the buildings tomorrow, and they're [00:11:40] not gonna use your water 'cause they can't, one, get enough GPUs to put in there, two, they can't get [00:11:45] electricity to, to supply them, and oh, they'll build a nuclear power plant on top of the data center, and all that [00:11:50] takes 15 years.
[00:11:50] Andrew: So good luck. This, this is the Silicon [00:11:55] Valley crazy optimism manifest at a level of society that I never thought would be [00:12:00] possible. You need that kind of level of optimism in a startup. You need that level of optimism to [00:12:05] change the world, but you also need healthy skepticism to balance it out.
[00:12:09] Andrew: And everybody's [00:12:10] just taken every-- these three CEOs', word for it[00:12:15]
[00:12:15] Nathan: Word. Yeah. I think the
[00:12:17] Nathan: key word you
[00:12:17] Nathan: Is healthy skepticism, right?
[00:12:19] Nathan: Yeah, there, [00:12:20] there are groups of people, right? But yeah, on at least with this topic, it's [00:12:25] the extreme optimism and the extreme pessimism, right? So you have the doomers, and then you have the, the [00:12:30] techno-optimists.
[00:12:31] Nathan: I, I'm not sure that's strictly a result of [00:12:35] the times that we're in, but we've seen it gradually moving that direction where w- at [00:12:40] least in the US, it is becoming more and more bipartisan. You're, you're either one or the other.
[00:12:44] Andrew: it,
[00:12:44] Andrew: Yeah [00:12:45] It's social media. We've talked about this incessantly, but it, it's just like the [00:12:50] level of curiosity of folks is so low, and the [00:12:55] curiosity has almost been reframed now to like instead of going like a root [00:13:00] cause analysis or apply critical thinking, it's become, "I'm going to just consume [00:13:05] all of the other predetermined viewpoints in my frame of view, and now I'm an [00:13:10] expert."
[00:13:10] Andrew: And it's like, no, you've just accumulated a bunch of points of view, which is valuable, but [00:13:15] you have to go to that next level. You have to evaluate those points of view to [00:13:20] understand if they're valid or not, or understand why there might be gaps. There is no perfect [00:13:25] viewpoint, and if you think you have a perfect viewpoint, you're wrong.
[00:13:28] Nathan: Hmm.
[00:13:28] Andrew: It's, it's just that, [00:13:30] So it, it's like there's always nuance, of course, but if you don't have critical thinking, you're never gonna get to nuance. [00:13:35] And it, it, the, just the amount of talking points that I hear yes, data centers can be cl- [00:13:40] catastrophically environmentally detrimental. AI is also-- [00:13:45] It does have promise to get to the point within a few [00:13:50] years where it is in an absolutely transformational technology unlike anything [00:13:55] we can imagine.
[00:13:56] Andrew: That's possible. I don't know that it's gonna happen. I'm definitely [00:14:00] not gonna feed the, the optimism. E-every time everybody predicts the future with certainty [00:14:05] they're wrong. It's just not how it works.
[00:14:07] Nathan: Right.
[00:14:08] Andrew: And with, with Moore's [00:14:10] Law and, and CPUs, we went through this. We're gonna double every year, and then [00:14:15] the '90s happened.
[00:14:16] Nathan: Hmm
[00:14:17] Andrew: And we got leapfrogs, but we didn't really have anything [00:14:20] transformational happen. And the Pentium 4 was transformational in terms of like [00:14:25] architecture, and it solved a lot of hard problems. I don't want to diminish it, but at the same time, it's no [00:14:30] one's life really radically changed that much until we started to get to true [00:14:35] like dual core computing.
[00:14:37] Andrew: And then we got all the software caught up like 10 years [00:14:40] later with actual like multitasking and multithreading capabilities
[00:14:43] Nathan: I th- I think that's an [00:14:45] interesting parallel to I guess timescale, right? You, you never really know how [00:14:50] much you're going to be able to build on top of the things that do come out, right? At that point in time when you [00:14:55] have something like the Pentium 4 or an AI data center, you only see the immediate [00:15:00] consequences, but you don't see that that actually contributed 15 years down the line to some [00:15:05] amazing growth or, or the other way around.
[00:15:06] Andrew: Well, yeah. When I remember when the Pentium [00:15:10] 4 this is nerdy now, but Pentium 4 came out, it was too hot, right? It was really [00:15:15] expensive.
[00:15:16] Nathan: I remember those. Yeah, yeah
[00:15:17] Andrew: And like hyper-threading kind of made [00:15:20] stuff better, but almost everything was single-threaded, and the Windows scheduler wasn't really [00:15:25] good enough to take advantage of it.
[00:15:26] Andrew: So like you got some boosts, but in a lot of, a lot of times you actually [00:15:30] got better benefit by turning off hyper-threading and just running things at a faster clock [00:15:35] rate. And it was a clock race, and then the clock race stopped because it was really hard and [00:15:40] we started to hit physics. So then they started saying instead of making one really, really [00:15:45] fast clock, what if we made multiple slightly slower clocks and put them together?"
[00:15:49] Andrew: [00:15:50] Okay, great. It's so much faster, but none of the software we use can take advantage of it so it [00:15:55] takes three to five years to... I remember having conversations with people,
[00:15:58] Nathan: thinking.
[00:15:59] Andrew: is so hard. It [00:16:00] doesn't make any
[00:16:00] Nathan: yeah
[00:16:01] Andrew: it," and blah, blah. And it is hard, and a lot of times it doesn't make sense.
[00:16:04] Andrew: But when [00:16:05] you can put 20 cores that are really fast into a CPU, it doesn't matter at a certain [00:16:10] point.
[00:16:10] Nathan: Yeah. This is super, super weird, but w- we were talking before this [00:16:15] about the-- we're thinking the same paradigm. There's one big thing versus a bunch of small [00:16:20] things.
[00:16:20] Nathan: Earlier, we were Yeah. about Fable.
[00:16:22] Andrew: Yeah it, is it better to have Fable, which [00:16:25] is a sledgehammer, or do you have just a ton of opus agents which are just smaller, slo- slightly worse?
[00:16:29] Nathan: If [00:16:30] you add them all together, maybe they do better.
[00:16:32] Andrew: Maybe we should talk about that, right? So it's like I've [00:16:35] been using Fable as I can, and we were talking about it. [00:16:40] My conclusion is that it, it seems like it's a better model in [00:16:45] every way. I'm sure it's not as good in some way, but everything I've ever [00:16:50] thrown at it seems to be better. It's also better at consuming tokens, and it's also better at like [00:16:55] token inflation, where it's using a lot more tokens for a similar task on a [00:17:00] different model.
[00:17:01] Andrew: It's also twice as expensive, and so like what you, what you were just saying, [00:17:05] like to break it down plainly for folks, was like if I'm using Fable for a task, I [00:17:10] would expect that task to be twice as good than using like [00:17:15] Opus or using an iterative loop with Opus and having Opus take a, a shot at it and [00:17:20] then an iterative improvement for two uses, 'cause it's twice as expensive.
[00:17:24] Andrew: And so you can [00:17:25] argue you get twice as many tokens. Tokens aren't tokens, they're not equal, but just to maybe [00:17:30] math it for some folks. So I, I am not [00:17:35] seeing a definitive so much better given the [00:17:40] cost
[00:17:40] Andrew: You know, if Opus gets stuck or like there's a hard problem, yes, it'll power through it in ways that [00:17:45] Opus tends to not. I, I did deploy it on something that Opus just failed at. I went [00:17:50] through three or four loops with Opus. It just could not figure it out. Took Fable two [00:17:55] times, so it was twice as expensive as-- Well, it was about probably equivalent cost 'cause it was [00:18:00] four to one.
[00:18:01] Andrew: But cool, so I just burned, what? I [00:18:05] don't know what the math comes down to, but call it 50% of my budget [00:18:10] on solving a problem where Fable... I didn't have Fable at the time, but how [00:18:15] do I know when to use Fable? Like they keep-- For the hardest tasks. Everything I'm doing is hard, [00:18:20] and I really don't see it frankly I don't wanna talk about the project thing that it's like [00:18:25] secret, but like
[00:18:25] Nathan: Oh, yeah, yeah, yeah.
[00:18:26] Andrew: I deployed it,
[00:18:27] Andrew: you
[00:18:28] Nathan: it done. Yeah
[00:18:29] Andrew: and it, it [00:18:30] made some interesting improvements, but it, or it made some suggestions for improvements for I'm [00:18:35] doing, I'm building an AI system and Fable had some [00:18:40] interesting analyses, but also did not go nearly as deep or to [00:18:45] the level that I was expecting for like model adjustments or optimizations [00:18:50] for model tooling, that kind of stuff.
[00:18:52] Andrew: It did ask me, it's "Are you cool with like [00:18:55] modifying the model or like training your own?" I was like, "Sure." And then it was like, "Oh, you should turn [00:19:00] off power saving on your CPU to make it..." yeah
[00:19:03] Nathan: Just let it [00:19:05] go.
[00:19:05] Andrew: Correct.
[00:19:06] Nathan: burn it all. Yeah.
[00:19:07] Andrew: And it's funny with the power limits, like [00:19:10] sometimes less power is more power or more speed depending on what you're doing, so it's not a yeah. Yeah [00:19:15] rule either.
[00:19:16] Andrew: Anyway, that's my rant on Fable. I don't know if you have anything to
[00:19:19] Nathan: No, I, [00:19:20] I, I-- we were talking about it right before, and I, I think that's... We've sorta seen it, I [00:19:25] think, come to this point where it was, like, each you may have heard the, [00:19:30] the term step changes. Anytime a new model came out, it was like a step change kind of thing, and I think [00:19:35] we've been at the point where these new model releases aren't quite step changes anymore.
[00:19:39] Nathan: Yes, there are some [00:19:40] problems that these new models are really, really good for like the other ones the older models [00:19:45] couldn't solve. But we probably saturated that point where a good majority of the [00:19:50] issues that we encounter, we wanna throw at these models now can be solved by these smaller models. That's [00:19:55] the, the space that we're, I think we're playing in now, and maybe that's the stop of [00:20:00] this, we'll see. This-- I was gonna say the stop of this pushing the, the frontier of intelligence, trying [00:20:05] to build the AGIs 'cause we're, we're at the limit seems like at least. I don't, I don't know [00:20:10] exactly what's going on in those frontier labs.
[00:20:12] Andrew: I don't think they do either
[00:20:14] Nathan: fair enough. [00:20:15] Fair enough.
[00:20:16] Andrew: The, But I-
[00:20:17] Andrew: I, I guess I'll qualify my statement [00:20:20] with the Fable stuff too. Given unlimited tokens, sure, I'm sure it's like [00:20:25] magical.
[00:20:25] Andrew: But
[00:20:26] Andrew: Am I gonna burn $2,000 every time I like open up a [00:20:30] prompt? No, that's not real. No company
[00:20:32] Andrew: is,
[00:20:33] Nathan: Yeah,
[00:20:34] Andrew: right? [00:20:35]
[00:20:35] Nathan: element into it
[00:20:35] Andrew: Yeah.
[00:20:36] Nathan: to make as much sense. Yeah.
[00:20:38] Andrew: reminds me of like mainframe time, right?
[00:20:39] Andrew: Where [00:20:40] it's like you, you kinda really do the work offline to use your 10 hours of [00:20:45] CPU to the most effective you can. And, and like you were talking about step modeling. I think this, [00:20:50] this is reminiscent of CPUs. I don't know why my mind's on that, but like the tick-tock cycle, which is [00:20:55] what Intel revolutionized after they
[00:20:57] Nathan: Ééé, é, é
[00:20:58] Andrew: they couldn't continuously [00:21:00] put out a much better thing, so they did a step improvement or like a, a iterative [00:21:05] improvement on a, a...
[00:21:07] Andrew: Trying to make this easy to understand. On a
[00:21:09] Andrew: [00:21:10] design. Yeah
[00:21:11] Nathan: Yeah, some... My brain farting on the word too. Initiative. Some [00:21:15] initiative,
[00:21:15] Andrew: it's the package and the die and all that stuff. But it, it for folks like who don't know what we're talking about or [00:21:20] don't care, like it, it, it was an iterative improvement. So it was a massive leap, [00:21:25] iterative improvement, massive leap, iterative improvement. And then they ran out of that, and then it became [00:21:30] tick-tock-tock, and it was highly controversial because it used to be basically every two years you were [00:21:35] getting a brand new CPU design, which tended to be a lot better, like 10, 15, 20, [00:21:40] 50% faster sometimes, and then Moore's Law kicked in.
[00:21:42] Andrew: And so-- Or Moore's Law was stalled, [00:21:45] I guess. Moore's Law says that's what'll happen, but it's like physics kicks, kicks in at some [00:21:50] point. And the same thing's happening with AI. Like we're in the early days, there's a lot of room [00:21:55] to go, there's a lot of juice to squeeze, so to speak. But like there's only so much blood you can squeeze out [00:22:00] of a turnip, and like all I'm seeing right now is like
[00:22:03] Nathan: That's a new, that's a new one. That's a new [00:22:05] one. Blow my... Say that
[00:22:06] Andrew: an old, old... Yeah, squeezing [00:22:10] blood out of a turnip.
[00:22:11] Andrew: Yeah That's the first I've heard of that. Okay. Sorry,
[00:22:14] Andrew: Burn up [00:22:15] juice. Yeah
[00:22:16] Nathan: I didn't know it was called blood.
[00:22:17] Andrew: I think it's kind of red color, so it's I [00:22:20] don't know
[00:22:20] Nathan: Ah, Okay.
[00:22:21] Andrew: idiom is. I'm sure there's some magical [00:22:25] 2,000-year-old thing about it, but, but yeah and it's the, the data center thing, [00:22:30] like, all of these claims are-- It's like [00:22:35] hypothetical compounded on hypotheticals,
[00:22:37] Nathan: Yeah. Right.
[00:22:38] Andrew: And sure, maybe it's all [00:22:40] real, maybe it's not, but it's, it's like you have so many levels of abstraction happening for all this [00:22:45] stuff. Yeah, if we just octupled the compute, it would be octuply better. And, [00:22:50] and the scaling law paper, the math's math, but so did Moore's law until it [00:22:55] didn't
[00:22:55] Andrew: Right.
[00:22:56] Andrew: And like physics happens and sh- shit's hard. [00:23:00] And when you start scaling things that much and you have like data center level compute, and then [00:23:05] you have multiple data center level compute, like there's all sorts of new challenges that interact that you have [00:23:10] to solve for. And sure, theoretically it might work, but the actual like physics and atoms and [00:23:15] people, it's not as easy.
[00:23:17] Andrew: And I'm still dubious. I don't think the transformer [00:23:20] architecture is gonna give us AGI. I think the bet is that if [00:23:25] we can blow enough compute on any problem, then the transformer architecture will work [00:23:30] well enough 'cause you can just run it 50 million times and it'll [00:23:35] statistically make the thing work the way that you want it to.
[00:23:39] Andrew: And I'm like, "That's [00:23:40] lazy."
[00:23:41] Nathan: Yeah, yeah.
[00:23:42] Nathan: But humans are naturally lazy, [00:23:45] right?
[00:23:45] Andrew: Yeah. So anyway, I wouldn't, A- a- and I've said before too, the number [00:23:50] one concern I have is not the resource stuff, it's the value that's being generated by the data [00:23:55] center needs to be returned to the society and the people
[00:23:58] Andrew: in the community, and also the [00:24:00] country. So if you open up an ice cream shop in some rural community somewhere, there's a couple [00:24:05] people who are gonna work there, you get some jobs, there's some tax base.
[00:24:08] Andrew: Maybe you have a [00:24:10] little bit of a tourism thing oh you're gonna attract people who don't normally stop in the town, right? [00:24:15] With a data center, you have a short-term employment boost sometimes. A lot of times they bring crews [00:24:20] from out of the area. And then you get a little bit of tax money for the land.
[00:24:24] Andrew: But you [00:24:25] don't actually get tax revenue based on the value that's generated. So if I scoop 50 ice [00:24:30] creams a day, and I you get a penny per ice cream, that's 50 cents per day in [00:24:35] tax revenue you get. Whereas if I push 50 bits across the server, I get [00:24:40] nothing
[00:24:40] Andrew: Right?
[00:24:40] Andrew: And I don't want to have a ton of taxes on everything.
[00:24:44] Andrew: I'm really [00:24:45] frustrated. I live in Washington State, and I'm very frustrated with the state of our taxes here. However, [00:24:50] we need to have a sane Tech tax that [00:24:55] taxes especially these hyperscalers and data center operators to return that value to [00:25:00] let not only the local municipality to make sure that they have water treatment systems for [00:25:05] all this excess water that these data centers are, are jettisoning out into the environment, so it can be [00:25:10] treated safely or recycled or electricity, and that the [00:25:15] society benefits from it, right?
[00:25:16] Andrew: It shouldn't just be the people with the capital who build the stuff [00:25:20] get to then ex- extract all the wealth out of the, the ecosystem and, and do whatever they want with it. [00:25:25] That wealth needs to be returned back, and I'm not talking about some stupid socialism, communism [00:25:30] nonsense. That's how our society works.
[00:25:32] Andrew: That's how it's worked the entire time. [00:25:35] That's what made America different than Europe, right? Is this feudalism, this idea that [00:25:40] I have power and wealth, and I'm just gonna continue to gain that, was something that was different here. We're like, [00:25:45] "Nah, we're gonna take that, and we're gonna make society better with the money and the value.
[00:25:49] Andrew: And [00:25:50] in return, we're gonna give you highly educated people. We're gonna give you roads. We're gonna give you safety. We're gonna give [00:25:55] you an electric grid." All these, what we call social services
[00:25:59] Andrew: Need to get [00:26:00] paid for. And at a certain point, there's a capital flip-over. We've [00:26:05] talked about this too, but there's so much capital at play that you literally can't-- You just [00:26:10] funnel more capital up, and that's a problem with capitalism, and the way that we've done that in the [00:26:15] past is taxes.
[00:26:16] Andrew: And there's other ways to do this too, but the problem is there is a tipping point. [00:26:20] And it's not hard work. The people who are saying it's hard work is it's [00:26:25] luck. It's skill, but it's also luck. To amass enough capital to tip the [00:26:30] scale to the point where you basically have an infinite money train is luck
[00:26:34] Nathan: Can you el- [00:26:35] elaborate a little bit?
[00:26:35] Andrew: Yeah,
[00:26:36] Andrew: multimillionaires and
[00:26:37] Andrew: billionaires. Yeah the, the [00:26:40] people who are at the h- the forefront of these conversations tend to be people who [00:26:45] have obscene amounts of wealth. And they see this as a way to eliminate [00:26:50] employees, which are expensive, so that they can get more wealth. And they see this as a way to eliminate jobs [00:26:55] and to create robot labor that they can have their private robot army do whatever they want [00:27:00] without questioning.
[00:27:01] Andrew: It's an-- It's, it's megalomania [00:27:05] manifest. And my argument is I don't know, maybe, maybe not, but at a [00:27:10] certain point of hundreds of millions or billions of dollars, you can't, as an [00:27:15] individual, return that wealth into the society. So like the primary economic principle at [00:27:20] play
[00:27:20] Andrew: is that I'm a business owner, I create value, which creates money, [00:27:25] and that usually manifests very easily through the banking system.
[00:27:29] Andrew: I make a million [00:27:30] dollars, I put it in the bank. The bank loans out that million dollars 10 to one, let's say, [00:27:35] and so now there's actually $10 million in the economy that people are using for [00:27:40] buying a house or starting a business or getting a car or repairing their roof or [00:27:45] whatever, right? And then they hire a roofer to come, and then the roofer [00:27:50] charges money and makes value and then puts their money in the bank, and then you get this like compounding m- [00:27:55] monetary flow.
[00:27:56] Andrew: This is a very simple, even though it's kinda complex, of how the [00:28:00] monetary system works. The problem is when there's one person or a very small amount of people who [00:28:05] make a disproportionate amount of money, that money doesn't actually really get returned back into like the, [00:28:10] the standard ecosystem,
[00:28:11] Andrew: and then it gets funneled into things like AI [00:28:15] ventures.
[00:28:15] Andrew: And then the people who benefit from the AI venture are a very small group of people who have very specialized [00:28:20] skills, and then that doesn't trickle back out. So it's the same way as other people have said, like [00:28:25] Bill Gates can only get so many haircuts, right? So if Bill Gates has all the money in the [00:28:30] world, Bill Gates can get, let's say one haircut an hour, let's say.
[00:28:34] Andrew: But [00:28:35] if you have 1,000 people who get 1,000 haircuts every hour, you now have [00:28:40] much more opportunity and you have much more wealth distribution, and the society actually gets [00:28:45] better because the people who cut hair now have kids and they send them to college and they're the next [00:28:50] Einstein or they're the next Mozart or...
[00:28:52] Andrew: So it's a, a fanning distribution problem, [00:28:55] and we haven't solved that with tech. And tech is very good at vertical monopolies, which is [00:29:00] how you accumulate this wealth very easily. We know this, but [00:29:05] we have allowed it to happen. And because it's an abstraction, because we can't see the oil, we [00:29:10] can't see the tires, we can't see the cars, the Henry Ford problem, it's [00:29:15] really hard for folks to make rules to break up those monopolies to [00:29:20] make that make more sense.
[00:29:21] Andrew: And now I'm just
[00:29:22] Nathan: Yeah. Well,
[00:29:24] Andrew: it's, that's the [00:29:25] core problem,
[00:29:25] Nathan: it's-
[00:29:26] Andrew: it, it's, it's and we need to fix it.
[00:29:29] Nathan: Yeah, we, we've [00:29:30] talked a little bit about this before, I think both on and offline, but
[00:29:33] Nathan: Some of the things [00:29:35] that, the topics that we, we bring up, mostly technology, to be honest, but the elements we [00:29:40] talk about creativity and art tho- those are a challenge to that system of how do you [00:29:45] actually fix that?
[00:29:45] Nathan: And the only challenge I would have for you on the technology front is, yes, [00:29:50] technology is good at monopolizing, it is good at verticalization, but one of the other things that we talk about a lot [00:29:55] with respect to technology is the way for it to democratize something, to bring the, the level or the [00:30:00] barrier of something more accessible to a lot of people, and that sort of evens the playing field. [00:30:05] It, it really, again, is, is more of a tool and it depends on how it's used and leveraged,
[00:30:09] Andrew: B-right, but it [00:30:10] only does that,
[00:30:11] Nathan: it in a certain way
[00:30:12] Andrew: it only does that if people have access [00:30:15] and if there is a competitive system that is [00:30:20] accessible without control. So if, if [00:30:25] the, what's a good example of this? The, an illustration of the point would've [00:30:30] been prohibition of alcohol. So they tried to gate the technology of making alcohol, [00:30:35] right?
[00:30:35] Andrew: It's been known by humans since time began, probably. We figured it out pretty [00:30:40] quick. For folks who don't know, you basically take some type of a grain and you make some sugar, and [00:30:45] the yeast eat it, and it produces alcohol or bad chemicals that'll kill you depending upon the [00:30:50] yeast
[00:30:50] Nathan: Moonshine. Yeah.
[00:30:51] Andrew: Yeah, I know it. And you get different types of alcohol, [00:30:55] some of which make you go blind, some of which don't.
[00:30:57] Andrew: So don't try this at home, but it's a relatively simple [00:31:00] process. It happens naturally in, in nature. If an apple falls off a tree and it [00:31:05] gets the right yeast in it, you get an alcoholic kind
[00:31:07] Andrew: of fruit juice. Yep. Yeah, fermentation [00:31:10] is, is, has been known. So they tried to gate the technology, and [00:31:15] during Prohibition to say nobody could do it, while other people had the technology.
[00:31:18] Andrew: Canada sent a [00:31:20] lot of whiskey to America, along with other countries. Oh, yeah. No, it was a huge problem. And people [00:31:25] were, were doing their own thing, right? Because they had access to stills or they could m-manufacture [00:31:30] a still to do it. And so then the, the Prohibition en-en-ended up [00:31:35] failing, but the solution to that was [00:31:40] making it more difficult to produce the substance or le- illegal.
[00:31:44] Andrew: [00:31:45] And so then you g- you conglomer- you conglomerated and effectively [00:31:50] monopolized the production of alcohol to a few different companies, which everybody probably knows their household [00:31:55] names now. brewing w- is like a different kinda thing, and like personal use [00:32:00] alcohol manufacturing is a thing.
[00:32:03] Andrew: But the point is that [00:32:05] if you c- if you didn't even have access to those chemicals or that process, you [00:32:10] couldn't do it, then you are in a, a technological monopoly, and that's what's happening with AI [00:32:15] is like I can't just go train my own model. I theoretically can, but I'm not gonna get anywhere [00:32:20] near, found- foundation model capability.
[00:32:23] Andrew: And what's hap- what's [00:32:25] already happened and what is continuing to happen is that they're [00:32:30] locking away that technology, and so the, the... it's a complete [00:32:35] asymmetric power problem. With every other technology in history, there's always [00:32:40] been some kinda counterbalance that fashions that other people... You invent a [00:32:45] spear then you double up the metal on your chest and now you're impenetrable, right?
[00:32:49] Andrew: [00:32:50] The, the bow was pretty easy to make. You got armor that [00:32:55] s- s- defended against it, and you crossbow. Crossbow was hard to make and it required engineering, so there [00:33:00] wasn't a lot of it. But you had countermeasures, right? Castles were [00:33:05] another technological that got f- made way to other different...[00:33:10]
[00:33:10] Andrew: But the problem with AI is because once you hit a certain level of proficiency, you can just [00:33:15] move so much faster and further than anyone else. If you don't have foundational access to it, you're [00:33:20] left behind or you're left with a much more inferior product And so I, I [00:33:25] 100% agree with what you're saying, but I think this is a very different situation.
[00:33:29] Andrew: And the [00:33:30] solution isn't to just have a prohibition on it. The solution is to make sure that everybody has access to [00:33:35] it. And the people who, and I'm done with my rant, I promise, and the people who keep [00:33:40] professing this, Sam Altman, Dario of Anthropic, and [00:33:45] Google, the CEO, none of these motherfuckers have put out an open model that's [00:33:50] worth I'm really mad about
[00:33:51] Andrew: it,
[00:33:52] Nathan: Okay. Yeah, that, that's more... Yeah. Okay.
[00:33:53] Andrew: right? ChatGPT[00:33:55]
[00:33:55] Andrew: did,
[00:33:55] Nathan: yeah.
[00:33:56] Andrew: ChatGPT OSG- GPT-OSS from a while
[00:33:58] Andrew: Yeah.
[00:33:59] Nathan: did something a [00:34:00] while ago. Gemma,
[00:34:00] Andrew: Yeah, Me- Meta was trying,
[00:34:02] Nathan: not Know.
[00:34:04] Nathan: Meta gave [00:34:05] up.
[00:34:05] Andrew: but, they were trying. Yeah, they've
[00:34:06] Nathan: they, I think they gave up. Yeah, they gave up on
[00:34:08] Nathan: it.
[00:34:09] Nathan: Um-
[00:34:09] Andrew: just put out [00:34:10] a, I think it was a tokenizer that looked pretty good. I haven't tried it, but I, I saw the, [00:34:15] the, the card for it.
[00:34:16] Andrew: But but the, the Chinese companies, what is it? [00:34:20] Z and,
[00:34:21] Nathan: Z.ai DeepSeek,
[00:34:23] Andrew: DeepSeek. What's the other [00:34:25] one? The m- the Minimax people. Kimi Mini- Minimax. I
[00:34:29] Andrew: can't keep [00:34:30] track of all these companies. Yeah. Qwen. yeah,
[00:34:31] Nathan: yeah. yeah. Although, yeah Gwen, yeah. I was gonna
[00:34:34] Andrew: so [00:34:35] they're open
[00:34:35] Nathan: on the US side, NVIDIA is, is a open source player, so
[00:34:39] Andrew: Yeah. [00:34:40] Thankfully, because they, they understand the problem if they get locked into a [00:34:45] couple companies, they don't have a good market to sell to. They wanna sell to
[00:34:47] Andrew: everyone, right? No, seriously.
[00:34:49] Andrew: And, and
[00:34:49] Andrew: [00:34:50] Jensen Huang has said that multiple times in, in non-direct ways. My-- But my point is the [00:34:55] Chinese models are open, number one, so you can run them.
[00:34:57] Andrew: But number two, the weights, the things that make the [00:35:00] models express or, quote unquote, "think" in certain ways are also largely [00:35:05] open. Not al- always. There's, yeah, there's some asterisks. But none of these other [00:35:10] companies have done that. And again, to democratize technology, where the [00:35:15] is it? They're like, "Oh, we need to be told to do it."
[00:35:18] Andrew: That. You, you can release it tomorrow. [00:35:20] The reason they're not is because they're it puts them behind potentially, [00:35:25] and, people get to see their special sauce, and it's pretty special right now for, for the most part. It's [00:35:30] pretty good. All the big three are all pretty good. But, that's not open.
[00:35:34] Andrew: [00:35:35] That's not transparent. That's not democratic. That's not whatever you label you wanna [00:35:40] throw on it. And I'm sick of the double speak at this point. And no, I don't think the Trump administration's gonna [00:35:45] fix it either. They keep saying they're going to, and I think they're gonna it up more, to be honest.
[00:35:48] Andrew: But we'll see. Maybe I'll be [00:35:50] wrong. I wanna make sure I hate on everyone in this segment of Andrew complains about something [00:35:55]
[00:35:55] Andrew: on a Thursday afternoon.
[00:35:56] Nathan: or dies. Everyone sucks or dies. See you later.
[00:35:59] Andrew: Hey, [00:36:00] Germany just put out a really cool model, like a German language model, but it... [00:36:05] Yeah, I sh-
[00:36:05] Andrew: I need to write this stuff down.
[00:36:07] Andrew: Yeah, yeah, no, it like literally I think dropped today. [00:36:10] I think it's a, get the numbers wrong, it's like a 32 parameter, but it's [00:36:15] like outpacing itself in terms of capability.
[00:36:18] Andrew: So like [00:36:20] promising, right? There's other people doing it, but the problem is like they're [00:36:25] every op- even the, the Nvidia stuff's like pretty good, but it's, it's I don't [00:36:30] know, I hate to quantify it, but it's 70% as good as like Foundation, and [00:36:35] it's really geared towards people like large enterprises, banks, people who are already [00:36:40] well-capitalized to
[00:36:41] Nathan: Sure. Yeah
[00:36:41] Andrew: be able to not get locked in sort of thing.
[00:36:44] Andrew: Which is great, [00:36:45] but like it doesn't really help small businesses, it doesn't really help people who [00:36:50] can't afford ChatGPT or Claude or Gemini, right?[00:36:55]
[00:36:55] Nathan: Yeah, yeah. I,
[00:36:57] Andrew: yeah. Anyway, I'm done.
[00:36:58] Nathan: no I, [00:37:00] I, it's a g- I think y- AI is, is uniquely positioned in, in this space [00:37:05] because I, I think it'd be wrong to say we haven't seen similar things before with [00:37:10] different types of technologies, right? Basic utilities and maybe more recently, like access to the [00:37:15] internet. There's still a battle going on with people not having access to the internet. [00:37:20] Although I think the difference with something like AI is the cost to provide the service is just [00:37:25] too great to be subsidized, and we're starting to see some of that happening now where the [00:37:30] subsidization era is pretty much over.
[00:37:31] Andrew: I, I, I wouldn't-- I don't know. Maybe. I doubt [00:37:35] it. They need, they need to go public first, and then they can't shut a bunch of customers, so they're in a little bit of a [00:37:40] quagmire of "We're subsidizing the out of all this stuff. We're gonna go public, [00:37:45] and then we're gonna have to release our numbers, and then if we shut half our customer base, our stock's gonna tank, and then we go [00:37:50] under."
[00:37:50] Andrew: So they gotta figure something out. I suspect the limits will just invisibly go [00:37:55] away, which is also super transparent and democratic. You have five bars of tokens [00:38:00] to use. How many tokens is that? It depends on your usage. Yeah, that's not an answer.
[00:38:04] Nathan: Hmm[00:38:05]
[00:38:05] Andrew: Uh, anyway, I, I do think AI is different, though.
[00:38:08] Andrew: It's, it's the [00:38:10] only technology that I'm aware of that we've really invented that has [00:38:15] the ability to create. I'm not saying it's sentient, I'm not saying it's alive, [00:38:20] but
[00:38:20] Nathan: Your, your terminology there was
[00:38:22] Andrew: I know, but, but what I mean-- I'll qualify what I [00:38:25] mean 'cause I don't have the word for it. It, it do- AI, AI does not create. It is a [00:38:30] statistical representation and output of, of an input.
[00:38:34] Andrew: [00:38:35] However, I can't go to the radio and have it create a software [00:38:40] program. I can't go to a radio and have it, reason a complex [00:38:45] formula for a business venture. I can't have it read 50 articles and [00:38:50] distill it down for me. And so there's an asymmetric
[00:38:54] Andrew: use of the [00:38:55] technology. And you can, you can talk about if it's good or valuable or whatever, but the point is, [00:39:00] is that the, the belief is that it will get better over time, which is probably true. [00:39:05] And if you don't have it now, you're already far behind and you're not learning those [00:39:10] skills that you need to advance forward.
[00:39:11] Andrew: So theoretically, AI will get better at being usable, [00:39:15] but like at that point, are you already done? I think, and that's the fear that a [00:39:20] lot of people are starting to express, is that this is one of the f- few [00:39:25] technologies that it's like if you don't have it, like gas in World War I, right? [00:39:30] If you have gas, you're like
[00:39:31] Nathan: on ar- you're dead on arrival kind
[00:39:33] Andrew: Yeah.
[00:39:33] Andrew: And if you don't have the [00:39:35] gas mask technology, you're dead. And then you kinda get it, but it's like the gas mask technology isn't [00:39:40] great, but now we're creating nerve agents for skin, and now we're just gonna kill literally everything that moves, and then we're [00:39:45] like, "Ah, maybe that's not a good idea."
[00:39:47] Andrew: So we, we came together and changed how we approach [00:39:50] it. There's a resistance to that today with AI and regula- and regulation [00:39:55] because we alone can't do it, because if we put a prohibition on it, China or [00:40:00] Europe or someone else is gonna potentially unlock this AGI, which I'll [00:40:05] qualify as AI that's capable of, of improving itself [00:40:10] in ways that humans can't in a
[00:40:12] Andrew: near linear [00:40:15] fashion.
[00:40:15] Andrew: And that's, that's the scary part that I think people are really getting to. Anyway, I
[00:40:19] Andrew: [00:40:20] promise... doomer stuff.
[00:40:21] Andrew: And I don't think we're anywhere near this, just to be really [00:40:25] clear. I also, I think we would be stupid to not keep pushing [00:40:30] it forward, but we have to figure out ways to make sure everybody is on [00:40:35] equal footing to approach the table and do [00:40:40] with it as they do.
[00:40:41] Andrew: I don't, I don't believe that we can just force it on [00:40:45] everybody, but everybody should have access to it, and your free 10 tokens or whatever you get [00:40:50] is not it. And we need a way to return that value, that the ge- the [00:40:55] value that is purportedly generated. Again, I'm really skeptical that there's a high ROI on any [00:41:00] of this stuff, to be honest.
[00:41:01] Andrew: I, I just don't see it for most cases. There's definitely [00:41:05] ROI. I'm making ROI with it. So I-- There, there is ROI to be had. It's just, I think, giving it [00:41:10] to somebody and saying, "Make ROI or make your job better," I don't, I don't think that's really it.
[00:41:14] Nathan: [00:41:15] Mm-hmm.
[00:41:15] Andrew: But-- And it's not redistribution. It's not [00:41:20] having American citizens owning the companies.
[00:41:23] Andrew: Not against ownership of the [00:41:25] companies, but what they're trying to do is, is, at least in OpenAI's instance, is they're trying to have the American [00:41:30] taxpayer own shares of the company, which effectively props up their [00:41:35] valuation, right? It's not to return value to the American population, it's to hold up their stock [00:41:40] price, and I don't think people are getting that.
[00:41:42] Andrew: We buried the lede, and you're gonna [00:41:45] be at Open Sauce.
[00:41:46] Nathan: Yeah, I
[00:41:46] Nathan: was gonna say, I'm I'm keeping an eye on the clock.
[00:41:49] Andrew: I [00:41:50] know.
[00:41:51] Nathan: Yeah, so the reason I'm here in San Francisco is, is there, there is a [00:41:55] conference or a, a convention, I should say, called Open Sauce. And it- it's really a [00:42:00] gathering of a bunch of creators who contribute in [00:42:05] the open sau- open source manner, and I will be exhibiting there.
[00:42:09] Nathan: One of the things that we [00:42:10] have talked about on this pod before, and it seems like it's been a long time 'cause I think the very first cast we [00:42:15] talked about this device, this Clockwork Orange helmet. that You had mentioned.
[00:42:19] Andrew: see it [00:42:20] live
[00:42:20] Nathan: Yeah. I, I will be an exhibitor at Open Sauce twenty twenty-six in San [00:42:25] Mateo. I'm here for the weekend trying to prepare this demonstration [00:42:30] that I have for, for the, the event itself.
[00:42:33] Nathan: Um, but...
[00:42:34] Andrew: soul [00:42:35] probably
[00:42:35] Nathan: No, no. For safety reasons, I am not going to have it actually active. [00:42:40] That was originally the plan, was to have it live, and you could try, try and test it out. Some [00:42:45] part of the back of my mind still wants to be able to do that, but I, I'm really focused on just [00:42:50] making sure I have something presentable so that I can, I can show it and talk about it.
[00:42:54] Nathan: [00:42:55] And I think the main reason I wanted to bring this device to something like Open [00:43:00] Sauce was Andrew and I have talked a bit about capabilities of open source and open [00:43:05] source science and what that might look like. For-- as a quick reminder, one of the driving [00:43:10] goals for this particular project is to present a stable and standard [00:43:15] platform for academics to be able to research the modality, which is transcranial [00:43:20] direct current stimulation, a bit more consistently. So everyone is doing their own experimentation, [00:43:25] everyone has their own parameters, and because of that, you have some academic papers saying, "Hey, this works [00:43:30] for this," and some academic papers saying, "Hey, this doesn't work for this." So just trying to get a, [00:43:35] a solid footing for which we make our conclusions is the idea.
[00:43:38] Nathan: And one way to start doing [00:43:40] that is to provide the hardware and the software to, one, digitize everything, but [00:43:45] also, take all the, the, the mental load off of the experimentation design because it's all [00:43:50] digitized and you can trust that it does what it says it's supposed to do.
[00:43:53] Andrew: So go see [00:43:55] Nathan and
[00:43:56] Nathan: you're, yeah, if you're in the area, yeah.
[00:43:57] Andrew: on poor Wi-Fi. But yeah I don't [00:44:00] know if there's tickets still available. It's, it's not really a plug other than it's a plug 'cause you'll be there. But [00:44:05] I'm excited to talk about this next week and get
[00:44:08] Andrew: your, [00:44:10] your analysis of, of kind of the, the data collection process and what you've learned and the, [00:44:15] the problems that you've encountered along the way and, and overcame.
[00:44:19] Nathan: [00:44:20] N-90% of it's gonna be no one knows what the heck is going on. My hair's on fire and we're just trying to figure it out. [00:44:25] But it'll be funny though. I'm excited to talk about it It's a typical weekend.
[00:44:29] Nathan: [00:44:30] Yes. Yeah, exactly. Typical weekend.
[00:44:32] Andrew: All right. We're over by a lot, [00:44:35] so
[00:44:35] Andrew: see you next week. Good luck. Go see Nathan. all [00:44:40] righty.