8,686 karma · joined January 29, 2009
What a great analogy. And IG/Tiktok reduce it into an even purer state - endless random videos, barely if at all connected, ephemeral stimulation you can't even remember 30 seconds after seeing it.
I know 50 year old adults who can spend entire hours just in this mesmerized state of flicking through these random feeds, seeing but not seeing, like some kind of drug induced hypnosis. I wonder what Postman would write today, were he still with us.
I'm no super-insider, I only hear industry scuttlebutt like everyone else, but I have about a 95% confidence that the last 18 months has just been about more and better, without any kind of real leap or breakthrough. More hardware, more data, better technique. Well, technique diffuses as people change companies, hardware can be built, and data can be gathered (or stolen!).
From my admittedly outsider perspective, the only years-long moat there is who has the most hardware. If you have the hardware, you can give away the compute to get the data (hello, subsidized subscriptions!). Technique can simply be hired. The only durable, multi-year advantage is the hardware.
So is that a moat? Sure, but it doesn't have a whole lot to do with the leading model companies of the moment. ASML is the real moat, and so it's ASML China is besieging, correctly (IMO) identifying that everything else can be caught up easily enough.
Check back in a few years...
Can't really think of anything else he's been right about, though. I don't think "right" is what he's going for anyway, it's all about that validation and a coherent, testable hypothesis takes a very distant second place.
It's pretty useless to compare raw FLOPS, but as a general hand-waving guesstimate, F@H is currently doing about 25 petaflops in a mix of FP16 and 32. AI usually trains at FP8, but to keep things fair the H100 is quoted at 60 FP64 teraflops per unit, so that's 12 FP64 exaflops given its 200k count.
So F@H at its peak did 2.43 exaflops@FP16/32. Colossus 1 does 12@FP64. These numbers are very hand-wavy, but I think the point is made.
By the way, I'm not trying to crap on F@H - I think it's an outstanding project and I've run it in the past. But a volunteer group simply cannot compete with well-funded, concentrated effort like what's going into AI.
Not sure what you are referring to, unless you don't think h100/h200/b200 are "AI hardware"
> Superpods aren't really power efficient
Maybe not compared to a specialized rig with multiple 4090s, but that is the best case for consumer hardware - the vast majority will be dramatically less efficient than that
Anyway, I agree the interconnect is by far the biggest obstacle and seems insurmountable, I should probably have led with that.
The far, FAR superior power efficiency means that even if you did harness every public GPU or GPU-like device on earth, you'd end up consuming so much excess electricity it would be cheaper on net to simply take the money that would have gone to the power bill and spend it on your own datacenter.
And even if electricity was free, having those GPUs spread over the world with internet-level latency will slow everything down by factors of thousands to millions - if it's feasible at all. Regardless, you're not getting fable-oss this decade, maybe even not this century.
It would be better for governments to buy and own their own datacenters, maybe as a coalition, and dedicate their operation to the public good. I believe that is what we actually have to do.
Very rough math like I said but I doubt it's directionally wrong.
And even if you did force literally everyone on earth with some sort of GPU to max it out 24/7 in service of an open source AI training enterprise - you would waste so much power trying to use that inefficient consumer hardware with the worst latency imaginable that it would be cheaper and faster to get everyone to instead chip in some cash to buy a datacenter with blackwell chips instead! So the idea has no legs whatsoever.
Your reflexively negative comments on anything relating to AI are as insight-free as they are numerous; it's all just vague shitting-on without even a hook or argument that could be engaged with and debated. It's pretty tiring, honestly. If you really think your point of view is valuable and others should pay attention to it, rather than just filtering it out like the trollish noise it usually is, why don't you put a little more effort in?
All that said, considering Anthropic's heavy-handed nerfing I'm not surprised Fable did poorly in a security-focussed benchmark. And this benchmark seems poor anyway - penalising a model for "cheating" by knowing the answer from its training data? That's not the model's fault, that's a lazy benchmark.
Michael Crichton said it best:
“Briefly stated, the Gell-Mann Amnesia effect is as follows. You open the newspaper to an article on some subject you know well. In Murray's case, physics. In mine, show business. You read the article and see the journalist has absolutely no understanding of either the facts or the issues. Often, the article is so wrong it actually presents the story backward—reversing cause and effect. I call these the "wet streets cause rain" stories. Paper's full of them.
In any case, you read with exasperation or amusement the multiple errors in a story, and then turn the page to national or international affairs, and read as if the rest of the newspaper was somehow more accurate about Palestine than the baloney you just read. You turn the page, and forget what you know.”
The Chinese models right now are in a weird spot. Compared to the frontiers, both their pre and post training is woeful - tiny, resource constrained in every dimension including human, slow. I'd compare it to OpenAI 5 years ago except I think even then OpenAI had way more!
But they "cheat" quite a lot in distillation and very benchmark-focussed RL and that's where you get this superficial quality in the leaderboards that doesn't match up when you go off-script. Arc is a great example in that it really belies an "inferior soul" at the heart of it all.
What gives me great hope though is that those same scaling laws that Altman and others have been hyping forever will absolutely kick in for the Chinese labs just as they did for the US ones, and I don't think anything can stop that process now. So they will catch up. It won't be tomorrow, but it's not going to be 10 years either. 3-5 would be my reasonably educated guess.
And the final risk, that China itself might try to restrict availability of the tsunami of GPU or other AI hardware it will inevitably produce - well, I just can't really imagine a country that has been configuring itself for the last 40 years as a single purpose export machine deciding that actually, no, it doesn't want to export something.
About the model restrictions - absolutely. I've been trying to do security research on my own software and the frontier models immediately get suspicious. I've been playing with the local ones much more this year basically because of this. They have deficiencies, for sure - they feel very "hollow" compared to the major labs. But I've talked to a lot of people, and the consensus is pretty clear - just a matter of time.
What bleeding? Anthropic wants as much of that "bleeding" as possible. The interaction data gathered from genuine human CC subscription usage of their models goes directly into their RL training, it's invaluable and they are more than happy to lose money on the inference to get it. That data is what xAI was recently willing to pay $10b to cursor to get.
They want you to use Claude Code. They hate other UI surfaces like OpenCode etc purely because they lose control over that data, so they're subsidizing the inference without getting what they actually want, the data (they still get some of it of course, but it's much less ergonomic for them. Those tools often abstract away the subagent calls, for example). OpenCode can collect that data themselves, so by allowing subscription there, Anthropic sees itself as subsidizing another org getting that data. Hard no.
And tools like OpenClaw are useless because they're mechanical and don't represent actual users interacting with the service - again, subsidizing but not getting the reward.
It's all very simple once you understand their motivations.
There is no secret sauce the US labs have that the Chinese ones don't, or won't have soon enough. Deepseek 4 and Kimi 2.5 are not quite Claude 4.5/GPT5.5 but there's no fundamental principle missing - they are strong evidence that there's no real advantage the "frontier" labs possess that isn't related to scale, which they will gain in time (if they even need to). The RL post-training techniques that work are widely known and easily copied. All Deepseek is really lacking is data, which they're getting - and the harder Anthropic/the USG makes it to access claude in china, the more of that precious data they'll get!
I used to sort of entertain the "fast take-off breakaway" scenario as being plausible but not really anymore. The only genuine moat the frontier labs have is their product take-up, which isn't nothing, far from it, but it's not some unbreakable technological wall. Too late guys - it might have been too late for quite some time.
> The thing is that even if I was wrong (I'm not) and AI was somehow helpful for software engineering (it isn't), I still wouldn't want to use it.
So even if you were wrong on the facts (you are) you still wouldn't change your mind? In other words, you're unreasonable and know you're unreasonable and think that's totally fine?
Well, cool. Next time, lead with that.
I understand the instinct to try to make a proprietary moat around it all but I think the pattern is useful and obvious enough that all big orgs will be doing something very similar within 5 years or so.
I welcome the experimentation, there will definitely be something new, but this ain't it. New primitives are needed, at a higher level of conceptualization, not merely a fancy new interface.
For reference, the huawei Ascend 950 that this thing runs on is supposed to be roughly comparable to nVidia's H100 from 2022. In other words, things are hotting up in the GPU war!
So, for me that was a production DB, fairly low usage but still. In the end I judged that it was going to take so much time to work around the outage that I just planned "maintenance downtime" myself and took it offline for a couple of hours until they finished.
Now, could I have architected better? No doubt. But it's a good reminder of what you're paying for and where the differences emerge in practise. AWS may be an overcomplicated, expensive mess - but I've literally never had any service of theirs go down for maintenance for any reason. Outages, sure, but taking things down for routine maintenace? Unheard of at AWS.
For me the formula is - I do whatever I want for my stuff, but if it's someone else's business, I use AWS.
Again, the article's numbers are likely a rather crude approximation, but taking 85% accuracy (claude 4.6) vs 90% (4.7) as inputs:
4.6 1 iteration 85%
4.7 1 iteration 90%
4.6 5 iterations 44.37%
4.7 5 iterations 59.85%
4.6 10 iterations 19.69%
4.7 10 iterations 34.87%
It's not like the military was specifically asking for mass surveillance, they just wanted "any legal use". Anthropic's made a lot of hay posturing as the moral defender here, but they would have known the military would never agree to their terms, which makes the whole thing smell like a bit of a PR stunt.
The supply chain risk designation is of course stupid and vindictive but that's more of an administration thing as far as I can tell.
The CEO of Anthropic, for example, has publicly stated multiple times that if their individual models were companies in and of themselves, they'd be profitable. I have no reason to think he's lying.
What, always? Like, literally 100% of the time if the merchant responds at all, they automatically win?
That's very hard to believe. I don't know Discover but I do know Visa and that's not how their system works at all.
The CFO just preferred Windows, that's it, I'd bet money on it.
The numbers aren't public but most guesses I've heard are that Anthropic's markup is around 50% on average, and that if considered in isolation, most models are profitable overall. The constant losses are instead due to training the next models, which will also eventually recoup but later, and forward capex investment.
This idea that big AI companies are normally and systematically selling inference at a loss as some kind of market share strategy is just not supported by the facts.
I an far, far less bullish than you on the timeframe. The vast amount of work is not even optimised for computers without AI. Much if not most bureaucratic process has stayed mostly the same for the best part of a hundred years.
It will change everything, but not in a single year!
It's not unreasonable to assume that without the ability to push haiku use aggresively for summarization, the average user in OC vs CC costs more.
My understanding is that the exact etymology is unknown. It's often written with the letter that references the tang dynasty, but the thing is there's no particular reason to think the Chinese introduced the style of cooking to Japan - although it is true that there was such a thing as fried chicken in 7th century China!
Another kanji-ization of the word uses the kara from karate (meaning air or empty, in karate it's "empty hand") and I find this equally plausible as karaage is fried with a very small amount of batter ("in air").
Either way they're both essentially competing "kanji backronyms" seeking to retcon an existing word as spoken; there's no real right or wrong answer.