fooker

33 ms·

901.
by fooker 1y ago
No, maybe you misunderstand the problem here? Given a specific program, yes it is measurable and predictable. A compiler has to deal with arbitrary programs, actually - small fragments of arbitrary programs at a time. Even if you could have
902.
by fooker 1y ago
It's a hard problem because the cost model is incredibly complicated. For example: you decide to unroll/inline/vectorize code has unpredictable effect on performance.
903.
by fooker 1y ago
Yes, the article is not self contained. Nothing is self contained, you are expected to look up memory tagging if you want the gory details of how other platforms implement it. If you want technical information about how the x86 ecosystem wi
904.
by fooker 1y ago
Given that this technique is used in production for all current gen Iphones should tell you this isn't vaporware.
905.
by fooker 1y ago
Lua does
906.
by fooker 1y ago
> China may never be able to export these chips While you type this, the rest of the world is already using Chinese cars, something that was unthinkable a year or two ago. The US has closed the market off from this for its auto industry
907.
by fooker 1y ago
> without AI We only call it AI until we understand it. Once we understand LLMs more and there's a new promising poorly understood technology, we'll call our current AI something more computer sciency
908.
by fooker 1y ago
If you Google zfs Linux data loss, you can find many posts about this. Including one lengthy discussion on HN. Also, you are not the typical user installing the OS from the default installer. I am not saying ZFS is bad, but not including it
909.
by fooker 1y ago
ZFS is pretty robust. ZFS on Linux is not, so much so that Oracle the company that owns it refuses to officially support it. It's a fairly well known issue. https://www.theregister.com/2018/04/10/zfs_on_l
910.
by fooker 1y ago
Hard drives are not getting better. Hard drives you can conveniently buy as a consumer - yes. There's a difference.
911.
by fooker 1y ago
> Because building a freezer that goes to -30 C is as cheap as going to -18 C. For small sizes, yes it is. But also, capex vs opex. Even if it's twice the cost, you only pay it once.
912.
by fooker 1y ago
> Tell me your wonderland where this has happened Plenty of grids had started paying consumers for excess solar generation exactly in this manner.
913.
by fooker 1y ago
I wonder if something like this happening finally pops the AI bubble.
914.
by fooker 1y ago
zfs on Linux has not been production ready for decades. People have lost data from it. There's no real reason to allow the default installer to do this. If you understand the risks, you can do it yourself.
915.
by fooker 1y ago
Given that you can buy a 48GB RTX6000 on newegg right now, yes?
916.
by fooker 1y ago
Barely.
917.
by fooker 1y ago
> used their own crawlers and indexes Not yet, but will eventually for sure. If you're an expert who has worked on Google search or something like that, this would b great time to start a company for this.
918.
by fooker 1y ago
Indexing is a fairly well understood technology. You could hire one or experts for this to be doable with a pretty good amount of scalability.
919.
by fooker 1y ago
Here's the interesting part - ranking matters when humans are looking at the results. For a bot with a large context window though, not so much.
920.
by fooker 1y ago
Writing a web crawler is not too complicated. I predict every 'AI' company will have a homegrown search engine in a few months to account for this. The way this would be publicly usable is through the new generation of 'AI&#x
921.
by fooker 1y ago
'American' companies have moved on to Fintech, hedge funds, real estate, and private equity. No big tech company now is run like an 'American' company. Take a look at the C suite and other leadership positions. Yes, AMD
922.
by fooker 1y ago
Because not everyone wants to pay for a subscription, certainly not the 5 billion Asians in any reasonable numbers. There's a reason Google and Facebook run the tech world as we know it.
923.
by fooker 1y ago
Interesting point. I can see this could turn out to be true. If we needed, for example, 1000 TWH to power AI for a huge drone swarm but could not do it while China could, this would be problematic. It requires a future where MAD with nuclea
924.
by fooker 1y ago
It's correlated to be data centers, not tied to. That's an important difference. We could easily build ..say.. 10 nuclear reactors and halve the utility bills with amortization.
925.
by fooker 1y ago
> Such a probe is quite conceivable, so why isn't there more evidence of such probes everywhere? Time, not space, is your answer here. Two reasons - (1) civilizations might not survive long enough to do this. (2) 13 billion years is
926.
by fooker 1y ago
> But games haven't stopped getting more demanding every year. This is not particularly true. Even top of the line AAA games make sure they can be played on the current generation consoles which have been around for the last N years
927.
by fooker 1y ago
When intel did it, the pitchforks came out. Nvidia seems to get a pass. Whys that?
928.
by fooker 1y ago
Is there publicly available evidence that TPU perf/$ is better than Blackwell ? I know it seems intuitively true but was surprised to not really find evidence for it.
929.
by fooker 1y ago
That is true for almost any cloud hardware.
930.
by fooker 1y ago
> And the Nvidia chips used to train it have barely retained any value either Oh, I'd love to get a cheap H100! Where can I find one? You'll find it costs almost as much used as it's new.

More ›