jmalicki

7 ms·

1.
by jmalicki 4d ago
> There is no similar currency that Nvidia can create ad infinitum. It can create obligations to provide future GPUs in return for present money. Yes at some point people might start to question, but what are the true hard limits there,
2.
by jmalicki 4d ago
The CHIPS Act in the US did set aside a reserve for older processes used for automobile, defense, and industrial semiconductors - but that's not gaming GPUs that used the previous process node.
3.
by jmalicki 4d ago
NVidia is limited by the number of chips they can produce. If you can fab 1000 chips, and can sell some for $500 and some for $80000 what are you going to do? The game GPU is at once profitable, but causes them to give up far more profits t
4.
by jmalicki 8d ago
> Do you really want the rust compiler to run at the speed of an llm? That... might actually be an improvement?
5.
by jmalicki 8d ago
> a Jeffreys prior will happen to make it satisfy the definition of a confidence interval as well Correct, which is why it's crazy to consider a Jeffrey's credible interval as valid, but not a confidence interval. Perhaps the c
6.
by jmalicki 8d ago
That's why this paper is important - it shows it isn't trained out. Leaving no other information in the model makes it clear what the biases are, and that the model is willing to make a biased decision. If you give it other unbi
7.
by jmalicki 11d ago
> Maybe their hardware is incapable I am unsure if it is incapable, but it sounds hard. They have tons of cores with 64k of SRAM each, and relatively slow paths in/out. On a GPU you can leave it in SRAM. On Cerebras, you have to se
8.
by jmalicki 12d ago
The insane valuations for selling a dream are what make VC worthwhile. TSLA would be worth crap if it were a private company giving off dividends. It is really truly about the insane valuations driven by collective delusion.
9.
by jmalicki 12d ago
> Something that is sustainable with a small team. These days, it can be sustainable for a tiny team, named Claude and Luna. Why doesn't every engineer have a side project or three for small market things of this caliber in 2026?
10.
by jmalicki 13d ago
I would maybe argue that Einstein was the most LLM-like of great thinkers. A lot of his great discoveries were mostly that he was very knowledgeable about the bleeding edge research in a number of disparate areas, and was able to have the a
11.
by jmalicki 13d ago
What if it can be Einstein, but can't draw a Pelican, write a solid college-level essay, or fold clothes? The ability to do a ton of book learning in training, and pull in tons of related context at once, is superhuman in some ways, bu
12.
by jmalicki 16d ago
It's no different than generally using AWS.
13.
by jmalicki 16d ago
The foundation of the businesses training AI models. So don't use them for inference.
14.
by jmalicki 16d ago
> It's quite shocking to me how many experienced, tech-savvy people, who used to care about cookies and ad tracking - are now willingly sending their business strategies, highly confidential contracts, and intimate personal issues t
15.
by jmalicki 16d ago
Are you training something so big you need that much unified RAM though? If you can fit it on a GPU, and especially for training, it is so much quicker than a Mac.
16.
by jmalicki 16d ago
It is a completely reasonable comparison for me as a consumer, since they're the costs and benefits that I'll actually get.
17.
by jmalicki 16d ago
> They also have enough power to negotiate contracts with strong privacy provisions. What privacy provisions would you want to add to AWS? Most of the reasonable strong privacy provisions you'd want are already there and/or av
18.
by jmalicki 16d ago
How would this allow compile-time resolution of virtuals that was not possible before? Gcc can already do link-time devirtualization in some cases.
19.
by jmalicki 18d ago
Interestingly embedded 32 bit Linux is not tier 1 or 2 though :)
20.
by jmalicki 18d ago
Sure, but I had to look: Tier1: Windows x64/i686, Linux x64/ARM gcc, Darwin/ARM Tier2: Linux x64/ARM w/ clang, Windows ARM, WASM, Darwin/x64 Tier3 is a pretty low support level, but tiers 1 and 2 is a pretty sh
21.
by jmalicki 18d ago
This is why I strongly like on-call. A lot of people hate it, but if you have a rotation, that will force this person to deal with what they create during their rotations, and the operational issues will be an explicit part of your job when
22.
by jmalicki 18d ago
There's also the executive's incentives, and "I'll be gone, you'll be gone": the payoff for that is very long, and it will be hard for them to take credit, and their own tenure might not be as long as that team
23.
by jmalicki 18d ago
Nothing here says the data isn't relational. It strongly disagrees, with reasons, why it's not better represented as relational. Personally I prefer the relational stance, and there are a lot of people who don't get it who s
24.
by jmalicki 18d ago
Some of that is that EU providers need to up their game here. Needing an EU native option is really the one and only reasonably objection I've heard against using LLMs from the cloud, the rest is tin-foil hat level unless you're a
25.
by jmalicki 19d ago
I am thinking the first place this appears is on GPUs and maybe Apple with unified memory, it is not going to be on PCs.
26.
by jmalicki 19d ago
The idea is to provide overarching guidance, but when there is devil in the details, leave that up to courts to consider when there is a real actual issue at play, rather than legislators in the ivory tower theorizing about how things might
27.
by jmalicki 19d ago
Since when has vagueness been considered a bad thing for statutes? It's the foundation of the US (and anglo) legal system and why courts exist, to fill in. And vague statutes that can be interpreted by courts has long been preferred
28.
by jmalicki 19d ago
From the opinion: "Because the definitions of event and occurrence do not resolve the issue before us, we must also consider the statutory context" They talk about this and the exact definitions for basically the entire opinion.
29.
by jmalicki 19d ago
The nice thing is that for GPUs you're rarely writing directly against the metal but instead pseudocode lowered by the driver; the driver could know to just lower it this way. On an NVidia GPU they control the whole stack and don&#x27
30.
by jmalicki 19d ago
Why would you want to read something someone was so sloppy at they didn't improve it with AI?

More ›