stymaar

14 ms·

301.
by stymaar 3mo ago
This blog post is pure slop, stealing from this one: https://climatecasino.substack.com/p/some-monsters-are-real the submission should be updated to link to the original instead.
302.
by stymaar 3mo ago
Elon's whole fortune derives from his ability to do illegal stuff without consequence, repeatedly.
303.
by stymaar 3mo ago
On a fast network, yes. If you want to create a distributed network over the internet, then it will.
304.
by stymaar 3mo ago
Indeed, I completely screwed my math up. Looks like 10am is too early in the morning for a Sunday. 33 tps max token generation speed would be for 10ms of network latency in the above example.
305.
by stymaar 3mo ago
The amount of data is only low for inference, not for training, and AFAIK DGX spark is supposed to be a researcher's machine that can do small-scale training.
306.
by stymaar 3mo ago
Throughput is not a problem as you just share relatively small vectors (a few kilobytes in size), the key issue is network latency .
307.
by stymaar 3mo ago
The catch is that the token generation speed is going to be limited by network latency, making it unbearably slow to run over the internet. It can be great on a local network though, especially if your workload is prefill-heavy (more text i
308.
by stymaar 3mo ago
Is it a fully custom inference engine or are you reusing parts of an existing stack? (llama.CPP, vLLM, etc.)
309.
by stymaar 3mo ago
> I note the lack of performance information. I can only imagine it's much, much, slower than any other way to run a larger model (including, e.g. using system RAM and streaming some stuff from disk) Not necessarily, and I suspect t
310.
by stymaar 3mo ago
Most people complaining about Rust being slow have never worked on anything close to a production app written in Rust. By the way, for most apps, rhe difference from an app we'd write in class is not as nearly as big as you'd expe
311.
by stymaar 3mo ago
AFAIK All of them have been retrofitted to take the lessons from TMI into account (I can't be sure about other countries, but in France it's definitely the case). And more importantly, the process around how you're supposed
312.
by stymaar 3mo ago
The lessons of TMI have been learned though, the accident has been thoroughly investigated and that's the reason why it's now being discussed in class.
313.
by stymaar 3mo ago
Sorry to break a myth, but you'll never hear someone about “THE RED BUTTON ” in a nuclear control room. There's way too much buttons that happens to be red for that. Nuclear operators are highly trained professionals (two years of
314.
by stymaar 3mo ago
Firefox mobile + ublock origin is the way.
315.
by stymaar 3mo ago
I'm pretty sure it's just the old photo look (plus the fact that in the current version, part of the space have been colonized by computers, which kind of ruins the mood).
316.
by stymaar 3mo ago
It's not specific to the Soviet world, any control room built before computers looks like that. The examples I'm familiar with is nuclear power plants from the 70s: - here's Bugey, the oldest active nuclear plant in France:
317.
by stymaar 3mo ago
Absolutely. We've just been better at engineering over time, and with synthetic fertilizer we gained access to a lot more of fertilizer than when we used manure. The same way, humans have engineered forests since prehistory, but there&
318.
by stymaar 3mo ago
> But maybe they could ask Claude how to train themselves to resist bullets as well? “unfortunately, my seven remaining comrades died in the process and I can't train anymore since there's no one to shoot at me”.
319.
by stymaar 3mo ago
> What do you mean by engineered? In a pasture for instance, grass can grow because the plant incorporate enough organic matter in the soil to be consumed by microorganisms that will in return fixate the nitrogen from the air into nitrat
320.
by stymaar 3mo ago
Public Choice theory is "a whole branch of political science" the same way "historic materialism" is though, with Buchanan instead of Marx, as it was created with the same kind of ideological motivations, with “state bad
321.
by stymaar 3mo ago
> The vast majority is grown on marginal land, just above pasture. I have no idea about the US, but in Europe it's absolutely not the case. We've replaced huge quantities of land that was twenty/thirty years ago dedicated
322.
by stymaar 3mo ago
As a friend of mine who also happens to be a math professor once said: mathematicians are like sculptors who marvel about the beauty of their creation, and are kind of disgusted when a physicist comes nearby and says “that's a cool ham
323.
by stymaar 3mo ago
It's wasteful in the sense that we are exploiting lots of land for the limited value it brings.
324.
by stymaar 3mo ago
> is it just as much a blatant jobs program as it seems? It's not a “job program” per se (these crops require basically no human work to do nowadays) but it's indeed a subvention program for farmers (and more importantly, land
325.
by stymaar 3mo ago
Related: https://en.wikipedia.org/wiki/Nixtamalization (Corn doesn't need special processing to be edible, but it does need special processing if you want to avoid dying from nutritional deficiency when having a c
326.
by stymaar 3mo ago
Well, at least not civilizations where dreams dry up.
327.
by stymaar 3mo ago
But “As grain gets more expensive” middle eastern countries (that rely almost entirely on import for their grain source) would start facing grain shortage (due to balance of payment issues) or at least severe deprivation of the poorer part
328.
by stymaar 3mo ago
> Today we have a huge buffer from the large use of grain to feed animals. This, plus the gigantic amount of agricultural land being used for biofuel production (almost as much as cattle food).
329.
by stymaar 3mo ago
See Zuck, the metaverse failed, but who said the mind control part couldn't work?
330.
by stymaar 3mo ago
So are opioids. Or how a very useful tool can become a public health catastrophe.

More ›