sarker
competency crisis actor
Suddenly I cannot remember the color of your eyes
Or the things we said as we stood together for the last time
User ID: 636
sarker
competency crisis actor
7hr ago
These were the guys who are worried about recursive self improvement and then we have Anthropic nerfing Fable's ML skills so it doesn't help competitors making AI, we have OpenAI guys on twitter saying 'GPT5.6 Sol did the post-training on GPT5.6 Luna'.
This was mostly adapting an existing config.
sarker
competency crisis actor
8hr ago
My last two healthcare visits (a GP and a physio) had a desk with one person behind it. I can confirm they were 20-60 (like, ah, most of the working population, I presume) but you didn't get the other particulars right.
sarker
competency crisis actor
8hr ago
My boomer father thought Claude was being politically correct by including "Negro performers" in a playlist for a party. I had to explain to him that many great musicians are black.
said hardware is not the most energy-efficient computation we can achieve
Certainly so far we have yet to beat the human brain's (general) intelligence per watt. I'm not aware of any LLM that can be run at all with the 20 watts drawn by the human brain.
I'll chalk up this line of argument as "extending lines on graphs." I don't fully discount it, it's useful as an outside view, but it seems inadequate as an inside view explanation.
sarker
competency crisis actor
17hr ago
Everybody loves Lord Gro!
His quick welcome wherever he turns is hilarious given that people either ignore his advice or take it and get doused with a literal bucket of shit.
The heroes are heroic the same way Achilles is heroic: brave, skilled, handsome, wealthy, of high status, of high renown and honour, fit to do deeds of daring and extreme feats, aristoi by nature and blood, far beyond the common herd. They don't have to be developed because they're archetypes. These are the Supermen of Nietzsche.
I think you do a disservice to Achilles and the characterization of the Greeks. It is not all sunshine and rainbows - I seem to recall an incident where Agamemnon takes Achilles's war bride and Achilles says he's going to take his ball and go home, and then later Agamemnon begs him to rejoin the fight with a very generous comp package and Achilles refuses out of spite. But I haven't read the Iliad so I won't argue too much.
The difference between Sanderson's functional if cardboard prose and his (very American) carefully worked-out magic system and Eddison's highly-coloured, vividly Jacobean-styled prose - !
We certainly live in a prose recession.
sarker
competency crisis actor
17hr ago
Sure. But the existence of a mountain peak does not imply that you can walk up there in shorts and flipflops. The origin of intelligence (and consciousness) in the brain are unclear. We know how many neurons are in a human brain, but we don't know if a neuron is the fundamental unit of cogitation. It might well be the case that Von Neumann level intelligence require simulating his brain in full fidelity. Even granting that such a thing is possible in theory, it is decidedly unclear that simulating the 10^26 atoms of his brain is practical.
sarker
competency crisis actor
17hr ago
It's hilarious that I'm apparently a skeptic despite saying right off the bat that I expect transformational impact on much of white collar work.
sarker
competency crisis actor
20hr ago
I'm saying even if we can only build a computer program that is approximately as smart as the smartest human ever
You are still assuming the conclusion. We have not built a computer program that is as capable as even a sub-median human in all domains, as far as I can tell, unless there is a program that can tie a shoelace and correctly tell me if I should drive to the car wash.
I don't mean this as a gotcha. LLMs are prone to certain cognitive biases that humans are not, and vice versa, and they are highly useful in many fields. But it's clear that the capabilities frontier is not uniform, far from it.
So what I'd ask you, as a full counter to my arguments, what upper limit or barrier is going to appear BEFORE we get to the point we've built something smarter than our whole species?
I don't know. All I know is that the current paradigm relies on massive amounts of artificially generated example problems with answers and I don't believe that all of human knowledge is amenable to such treatment. So far I have not seen any reason to believe that actually general, rather than spiky, superintelligence is imminent. And the imminence is, again, really the key question that's motivating all this.
sarker
competency crisis actor
20hr ago
Left-aligned organizations don't fear that, and they have gotten away with fedposting
Perhaps I missed it, but I didn't see that in your post, unless we're counting death celebration as fedposting now.
The moderators here have included far less overt advocacy than "His killer committed a just act." as fedposting.
Do you have an example?
sarker
competency crisis actor
20hr ago
But LLMs are getting freakishly good at things they haven't been specifically trained on. Their intelligence does generalize.
Such as?
Perhaps we only need to RL them in a few more domains to clinch the rest of generalized superintelligence. E.g. you can have them pilot robots and put them in virtual environments and RL fast them there, or real environments like an academy (a warehouse) a bit less fast.
There's been impressive seeming advances in robotics, though I'm not keeping up too closely. I don't see the connection between operating a warehouse and superintelligence though. Certainly the humans operating the warehouse are not superintelligent.
sarker
competency crisis actor
20hr ago
We should, in principle, be able to build a simulated Von Neumann that is ~as smart as he was.
This is basically assuming the conclusion though. Even granting this for the sake of argument, it doesn't mean that we'll be able to build such a simulation in the next 10 years rather than in ten thousand.
sarker
competency crisis actor
20hr ago
I don't think there's strong evidence against these but I don't think there's strong evidence for these either. Certainly LLMs are not more efficient than the human brain.
The conceit is that there is no such task, and so its only a matter of time, and adding capabilities to existing models, that the human capabilities are exceeded on all fronts.
Could be. But this isn't an argument for short timelines, which is implicitly what we're discussing here.
If the resulting entity is able to do self-improvement, it by definition will do so faster and more efficiently than humanity can track.
Only if, with self-improvement, it actually improves things that aren't suitable for RL environments with massive amounts of data. So far we are very much in the "lumpy capabilities" regime.
sarker
competency crisis actor
21hr ago
I don't really understand what you are trying to say here.
First you say that fedposting is allowed only against figures on the right without lawfare. Then you provide a long post with examples of people saying kirk deserved to be shot and then say that no discord channels were shut down for this sort of behavior.
However, it's quite clear that saying that someone deserved to die is not fedposting. So the connection is unclear to me.
sarker
competency crisis actor
21hr ago
I don't get it. Is fedposting against the right allowed on this forum?
sarker
competency crisis actor
21hr ago
Maybe someone here can help me with this.
What is the bull case, beyond drawing lines on a graph, for AI achieving superhuman, or even human, performance on tasks that are not quickly verifiable?
AI is quite clearly superhuman at self-contained programming problems. I haven't tried Fable, but I suspect that superhuman open ended software engineering is not far away, though I suspect that humans will have a role in architecture and problem setting as opposed to problem solving for some time more. I expect hardware work will also quickly go down this path, at least to some extent, and really anything that can be RLVR'd. That's enough to account for a huge portion of white collar work and carries serious cyber security risks. Both of those will have serious consequences, politically and militarily.
I am not convinced that AI is improving at anything like this rate for things that can't be RLVR'd, I.e. stuff where you can't generate enormous amounts of useful training data with an answer key. Radiologists continue to do just fine for themselves despite repeated promises of doom. I'm sure someone will chime in to say that the radiologists are there for liability reasons, but it's not as if they are now just hitting thumbs up/thumbs down on AI decisions all day.
Partly this is a sample efficiency question - there simply might not be enough data for them to learn this stuff to human level, and architectural advances that improve sample efficiency may lead to huge gains in quality. But it's not clear to me why people expect this to happen.
sarker
competency crisis actor
22hr ago
Indeed, you can run inference on the ground. Exactly my point.
sarker
competency crisis actor
1d ago
sarker
competency crisis actor
1d ago
I am indeed aware of that. There's advantages to having the compute collocated, which is why StarCloud is doing it that way.
sarker
competency crisis actor
1d ago
Of course you can make the radiator smaller by making the satellite smaller. But you lose any economies of scale by having a big cluster of compute, which is presumably why Starcloud is targeting a massive DC.
My point is not that you cannot run a computer in space, obviously. My point is that small DCs are unlikely to be useful and large DCs are unlikely to be feasible.
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 hope I didn't mislead anyone into believing that 2000 square meters is a megastructure. Nevertheless, most people have never seen an Ascend 950 so I don't think that helps contextualize anything for anyone. 2000 sq m is fairly large for a space radiator - the ISS has only about 400 sq m.
Or you can simply launch a little higher. No matter how you cut it, it's all ultimately about mass.
Perhaps. And yet, Starcloud plans to operate in LEO. I assume they aren't totally retarded and have thought through the choice of orbit. 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. Either the people working on this are smart and have chosen the best parameters for this, or they are stupid to the point that the internet peanut gallery can do better and therefore aren't going to succeed. You must pick one.
Neither solar panels nor radiators lose function quickly from random point damage.
If you're pumping coolant through a tube that's open to vacuum, you're going to have some problems.
Since you dislike X, I'll cite it again. NVIDIA CEO JENSEN HUANG: 1GW AI FACTORY ON NVIDIA ARCHITECTURE COULD COST NEARLY $100 BILLION
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 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.
Space based DCs also fail the "not currently happening" test, so this part is a wash.
sarker
competency crisis actor
2d ago
Numbers presented without showing the math can be dismissed just as readily.
As explained, it's a simple application of the Stefan-Boltzmann law.
2000 sq m is for a small DC. StarCloud proposes a radiator of nearly 8 sq km.
sarker
competency crisis actor
2d ago
Ascend 950 SuperPod has an area of 1000 square meters for the actual scale-up compute unit that works as one GPU.
Uh, okay. Manhattan has an area of 20 square miles. What's the relevance to space radiators?
«Redwire Q-Rad Deployable Radiator (commercial, TRL 5-6): 3.5–4.9 kg/m² areal density. Source: Redwire radiator datasheet lists; brackets our StarThink V1 assumption as plausible near-term path.» Let's say 4. This is just 8 fucking tons. This is peanuts. At $100/kg it'd be merely $800000 for delivery, 8% of Starship capacity, or four more big Starlinks. 1 megawatt of compute costs… let's see, a modern-ish GPU that draws 1 kW in a rack can go for $20K at least, and actually we'll see prices creep towards $50K. Well there you have it, $20 million as the floor (and GPUs are just ≈40% of BOM). 1 megawatt is 8760 MWh/year. Google tells me wholesale electricity in the US is like $45. Almost $400K a year of free power. None of this matters of course, when inference margins are >80% even with hardware depreciation, and all that matters is deploying as fast as possible, as much as possible.
Hmm. What is StarThink?
I think it's quite clear that mass is not the concern I raised wrt the size of the radiators, which is why my post did not say anything about mass. Huge radiators cause drag and are vulnerable to micrometeorites. Redundant cooling loops can mitigate micrometeorites, but then you need bigger radiators.
But since you brought it up - the radiator you posted is solid state, SpaceX says they will use circulating liquid radiators, same for starcloud. I expect this is because circulating liquid radiators scale more than passive radiators like the one you are looking at. Just a guess on my part, though.
Now, StarCloud says each data center will be 5GW. They are cagey about how exactly they will dissipate 5GW, but with a simple water/glycol loop we are looking at 50,000 kg/s of flow which is massive. All those pumps will of course also need to be cooled.
On top of that, the radiators need to be filled with coolant. Back of the envelope math, again for a simple coolant loop, suggests perhaps 100,000 tons of coolant in the radiator, so we're looking at $10B just to get the coolant up there across hundreds of launches and nothing else. They do not go into the math of this in their whitepaper, probably because single phase liquid coolant is totally infeasible.
You don't know the relevant numbers in any of the involved verticals, and for some reason (unfathomable to me) you want to believe that the numbers support your (quixotic but perplexingly popular) case against compute in space. They don't.
Tedious bluster. Please save it for Twitter.
I still don't understand. Your original claim is "Amadan has specifically and repeatedly noted concerns about a possible warrant to Zorba as a major motivating factor [to moderate fedposting]" and that by "fedposting" they include things far less overt than "he deserved to die." As evidence, you cite Amadan saying that TW left the motte because of fedposting, and, separately, pointing out that TW linked to a certain comment in his farewell. However, nobody is saying that that comment was itself a fedpost, and in fact it was not moderated at the time.
At best you can say that Amadan said that TW said that FC expressed a desire for violent conflict. But there's no mod censure to speak of here.
More options
Context Copy link