DaseindustriesLtd
late version of a small language model
Tell me about it.
User ID: 745
DaseindustriesLtd
late version of a small language model
4d ago
Personally, I just don't share the optimism of these guys in either direction.
I don't know what you mean by optimism. They expect Plan D to be the most likely path, and one of the crappier ones. Plan A is just an incredibly weird scenario they've built around their assumptions, that prevents both violent ASI takeover and Total China Victory. Given that alignment is probably easy and introduces modest capability tax, their whole paradigm is wrong.
Full automation of labor is a technical reality within 10-15 years.
DaseindustriesLtd
late version of a small language model
5d ago
Optimus is great. Very good hands (some Chinese stole them), powerful, iterates quickly, and more importantly I can trust Elon to mass produce it, as he mass produces lots of things. Sure it's a bit quaint compared to the Chinese robotic supply chain and scaling potential of Unitree and UBTech and others. But regulatory barriers will all but ensure that Western markets heavily go to Elon.
The problem with Optimus and with Tesla taxis is the same: it's a bet on the exponential, and you don't know your exact location. Elon's theory of victory for FSD is that good enough AI will make do with human-level sensorium; arguments about lidars being expensive are of course nonsense, the costs of lidars can fall like costs of any other component. He's obviously correct on the substance; the question is what does it take for "good enough", how much more data, pretraining and onboard processing? He keeps discovering that the answer is "more than you have". But at some point, very likely it just works and Waymos become overengineered toys.
Well, but if it's not time for robots, why is he acting like this is the very thing that will take the entire company to an entirely new level? Transform human society, even. Shouldn't his car company focus on cars, and leave humanoid robots as a niche R&D project?
I can't muster the outrage. His corporate governance experiments have trivial explanations, and he'll have the cash to burn on it all.
but V3 will be able to take 100t
I think you miss other variables changing.
and how far they already got with their own rockets of similar class
The problem is they haven't gotten far. If rocket reliability requires exploiting Wright's law, Elon is very much ahead.
DaseindustriesLtd
late version of a small language model
6d ago
We have an ASI benchmark, courtesy of ByteDance Seed. Or rather, a framework for one. https://edge-bench.org/ has no ceiling.
Though what does it matter? The steam hammer won.
DaseindustriesLtd
late version of a small language model
6d ago
It's difficult to have a discussion about this when you ignore the details from the actual proposals in favor of advocating for stuff they aren't doing when it's convenient
You know what, fair enough. Let's ignore Starcloud since this is primarily about SpaceX. They've just issued a concrete design: Starmind
• 150 kW peak compute payload
• 120 kW average compute payload
• 70 kW per ton
• Wingspan: 70 meters
• Deployed height: 20 meters
• 110 m² deployable liquid radiator
• Redundant pumping loops
• Integrated micrometeoroid shielding
• 150 kW solar array
• 250 W/m²
• High-speed laser links interconnect satellites and beam AI results back to Earth through Starlink. Low-latency, high-bandwidth connection
• SpaceX-manufactured solar technology from Bastrop, Texas
So, that's 917 square meters of radiator per 1 (sustained) megawatt, and more importantly 70 kW of capacity per ton, at SSO. I see Starship has the theoretical capacity of 40-60 tons to SSO, let's say 50. At, say, $200/kg that's $8M to deliver 2.8 MW of compute. As per Jensen, 1 MW can go for $100M. There's plenty of slack in this. Even if Jensen is off by an order of magnitude, the "getting it into space" part is almost a rounding error and can make straightforward sense given terrestial/political constraints.
DaseindustriesLtd
late version of a small language model
6d ago
Elon himself is very well aware of this. I sometimes say that he's spiritually Chinese, which is lost on his greatest admirers in the West, who imagine his success is due to some brilliant insights. No, it's similar to what guys like Lei Jun do, just with American capital scale and more chutzpah. It's maybe the most potent recipe there exists.
Fundamentally, manufacturing is underrated and design is overrated. So people generally think that there's like this Eureka moment of you come up with this idea and and that's it, now it's good. But as good as a design is, it could literally be that a thousand percent, maybe ten thousand percent more work that goes into the production system of the thing itself. So, how much effort we've put into say designing Rapor, versus the manufacturing system? Uh it's ten to a hundred times more effort to design manufacturing systems than the engine.
– Even for Raptor?!
– Oh yeah, absolutely ESPECIALLY for Raptor. The amount of effort that goes into the design rounds down to zero. Relative to the amount of effort that goes into manufacturing. Yeah. If this is not true, great, I'd like uh a thousand Raptors. Uh oh we can't make them? Oh alright. This is like just very fundamentally underappreciated. If people have not been in manufacturing, especially manufacturing of something that's uh relatively new, then they don't understand and they they think the design is the hard part and they think production is like the copier or something like that. This isn't said enough, I'm trying to correct the misconception that design is the hard part, it is not the hard part. Uh, there's been lots of great rocket engine designs. I've spent a lot of time looking at the Russian rocket engine designs. There's some amazing Russian rocket designs. They've been doing stage combustion for a long time, yeah. And they've done I don't know, hundreds of different designs. So the the hard thing is not any design of staged combustion. This has been done. Yeah. Um now admittedly our is higher pressure than before and it is a full flow combustion. But these are, that that's a relatively minor increment relative to what the Russians have already done. Right. What is super hard about Raptor is uh, how do we make a Raptor where the cost per ton of thrust is under a thousand dollars.
DaseindustriesLtd
late version of a small language model
6d ago
I don't challenge your reasoning for making resolvable bets, my problem is that they don't have much relation to the interesting question. This is the usual forecasting problem.
Your specific arguments for why Starship can work all sound reasonable to me, but they don't sound different to me from arguments for why Cybertruck could be a good truck, why FSD could drive safer than human drivers, why optimus could be a great humanoid robot, etc.
All of these efforts being meh (so far) is not very informative. Cybertruck is just a goofy car, there's only so much you can achieve by making a big electric pickup with edgy body panels. It adds very little to Tesla's current position. Car people are somewhat insane in paying so much attention to car models. FSD works, Waymo is reportedly great, so Tesla robotaxi also could work. Optimus is a legitimately good robot, it's just not the time for robots yet, and China is way ahead of Elon on the entire robot supply chain except high-end chips (both the brain and external compute). Starship is a categorical breakthrough in space logistics, which is the one area where Elon is far ahead of the competition already. There is no way for others to overtake him on any reasonable timeline.
Didn't he buy Cursor, and these guys were the ones who figured it out?
He did, and no doubt their data has contributed a lot. But the base model is in-house, and I see that RL was done on xAI's stack. This is impressive because the original xAI team has completely fallen apart, there was the impression that xAI has become a mere compute provider for Anthropic. He has crashed and rebuilt a near-frontier lab from the ruins. This suggests at the very least good capability for delegation outside the hardware domain.
DaseindustriesLtd
late version of a small language model
6d ago
I don't really understand what drives a man to repost second hand all caps claims. I'm not even saying that he didn't say this, but surely you must understand that this is simply not convincing to anyone?
I was too lazy to de-caps it, and I hope that people of this forum will find the issue of the costs of 1 gigawatt of capacity on Earth more salient than the funny detail about all caps.
Problems of space compute have straightforward engineering solutions, the costs of which can be estimated. Whether these solutions are worth the cost depends on the costs of building the same capacity on Earth. So arguments about radiator area, micrometeorite damage or coolant mass are kind of… weightless unless grounded in comparison to the baseline.
DaseindustriesLtd
late version of a small language model
6d ago
The orbital centers may look cheaper in back-of-napkin calculations right now, but only because battery prices haven't yet crashed as far as solar panel prices have and chip prices are so high that you want to run everything on a 100% duty cycle
Solar prices are already creeping up as China has ended subsidies and the global demand is surging. The Chinese will do all they can but at the end of the day PV panels have scarce physical inputs (like silver). The lowest realistic price for batteries that I've seen was something like $15/kWh (if Sodium-Ion works out at scale). That's about $300000 for year-round battery+solar 1MWh supply (given seasonality and losses), probably more. Plus immensely more costly solar installation (lower area efficiency, overbuilding due to day-night and seasonal cycles, weather protection, land)…
Might as well just yeet it into orbit.
DaseindustriesLtd
late version of a small language model
6d ago
Uh, okay. Manhattan has an area of 20 square miles. What's the relevance to space radiators?
It matters because it deflates the context-free appeal to "omg 2000 square meters". Ok. 2000 square meters is just 40 * 50 meters. Is this supposed to be a lot?
I think it's quite clear that mass is not the concern I raised wrt the size of the radiators
It's not clear, because mass is the only interesting concern there is. Drag, too, is an issue of mass (for ion thruster fuel). Your link says: "And that’s the best-case scenario. Additional problems are hidden in the low Earth orbit environment itself. Space exposes radiators and their coatings to a chemically hostile brew of ultraviolet light and atomic oxygen, quite the opposite of a clean-room environment. Over a LEO satellite’s typical 5-year lifespan, these elements degrade the radiator’s surface properties and lower its ability to shed heat. … Including this degradation in the model reveals that as the radiator degrades from a “fresh” state to an “end-of-life” state, the physics demands a further penalty. To maintain that same 60 °C operating temperature for the GPU chips, the required surface area jumps from about 1.4 square meters per chip to nearly 2.0 square meters. In other words, the physics tax rises by 40 percent. Therefore, you must launch at least 40 percent more radiator mass, endure higher atmospheric drag, and sacrifice valuable launch volume just to survive the degradation of the thermal coating."
Or you can simply launch a little higher. No matter how you cut it, it's all ultimately about mass.
Huge radiators cause drag and are vulnerable to micrometeorites
Huge solar panels cause drag and are also vulnerable to micrometeorites. This issue, just like the radiator issue, is negligible. Neither solar panels nor radiators lose function quickly from random point damage. At, say, 500 km the lifetime of an inference node with several thousand square meters of total area can be a decade. How much economic value does a decade of compute with free power provide? That, cost per kilogram to orbit, and costs of hardware are all that matters.
Since you dislike X, I'll cite it again. NVIDIA CEO JENSEN HUANG: 1GW AI FACTORY ON NVIDIA ARCHITECTURE COULD COST NEARLY $100 BILLION
So, maybe $450B for that 5 GW you talked about. Cooling alone is likely a fifth of that. I guess popular reporting can create the impression that Americans are actually standing up tens of gigawatts of capacity without problem, like so much coal plants in China. This is not, in fact, happening. Most Blackwell compute is still not operational. All this space math only matters in relations to costs on Earth.
This is an incredible shitshow and, yes, seemingly the standard for modern wars. I've said before that the MOU is a sign of the US accepting the failure of Epic Fury and military defeat and cutting losses (contra my initial expectations; tbh that was impressive, if anything – Russia also miscalculated, but can't even cut losses). But well, nothing (except self-respect and reason, perhaps) could prevent the US from retroactively revoking this acceptance and just relaunching the war. Opening the Strait by force is going to be as hard as before, the oil supply crunch won't get any better, interceptors and standoff munitions didn't magically regenerate in the meantime, but it didn't become any easier for Iran either. Will the Regime finally Collapse under the pressure? Doubt. Will it accept less favorable terms of peace, now that American resolve is reaffirmed? More plausible but I think their payoff matrix is unchanged (they need to secure and fund post-war military deterrence in the presence of Israel, they can't trust the US or Israel), and that dominates their analysis. Will we see escalation, maybe even ground operation? Seems remote…
I suspect one big factor of why Trump feels like he can afford this (and why he doubted that enough to authorize the MOU before) is that the oil demand is proving to be sustainably low, and that's mostly due to China sharply dropping their oil imports and thus not competing for a volume to the tune of 5.5-6 millions of barrels per day, roughly 2 Japans' worth, vastly more than what they were getting from Iran and Venezuela combined. For all of Trump's gratitude, this is not charity from his "friend" Xi, of course. The best analysis of what's happening that I've seen is here, in short:
They may also be releasing their SPRs, but that's not clear, nor is the volume of this buffer; I've seen numbers from 1.2 to 2 billion barrels. Combined with domestic production of ≈4.3mbd and some 6.4mbd of remaining imports, they can presumably keep this up for years. Their strategic objective is removing oil import dependence anyway (much to the chargin of Malacca blockade enthusiasts). There may be backchannel verbal agreements too. So Trump can assume that there won't be a sudden spike to $150+ per barrel, gas pump voter outrage and other externalities of the closed Strait.
Another factor is that there's a planetary bull run driven by AI and upstream HPC hardware supply chain, where oil-importing Asians are all winning massively, the US is getting huge capital inflows, and that's papering over bad war news. American dominance in AI, demonstrated with Claude Fable release drama, is also being rapidly converted to diplomatic leverage over the squeezed NATO allies.
So if Trump's thinking is dominated by things like midterms, he might just figure it's better not to be a loser.
Still, energy inflation is really bad, already 15.8% year-on-year in OECD. The war will continue to function as a tax on global growth, worse for some (Europe) than others, for the foreseeable future.
More options
Context Copy link