jmalicki

4 ms·

1.
by jmalicki 6d ago
> Labs will try to filter it out, but it will appear in web search results too. Sounds like religious discrimination.
2.
by jmalicki 7d ago
If it's closed source, how do you know it's not technology given by aliens from the 43rd dimension running on quantum computers enabled by discovering that P=NP and finding a linear time reduction from NP to P? Occam's razor
3.
by jmalicki 7d ago
For the Jev use case for LLMs, do you mean having the LLM produce a probability as text ?
4.
by jmalicki 7d ago
That goes all the way back to at least to Stein's Paradox in 1955, sadly too few people get educated about Statistics and keep thinking specialized models will necessarily be better. If you want to estimate the batting averages of 3 M
5.
by jmalicki 11d ago
Nothing in your reply gets at the connection to out of sample data?
6.
by jmalicki 11d ago
Okay, and that's all in-sample, which is the entire point, it won't necessarily hold out of sample. E.g. over-fitting.
7.
by jmalicki 11d ago
You are saying something interesting, but talking like Grok and skipping a lot of the details, without any references to common check-in points like terminology or specific studies. > and concentrate the likelihood around the zero loss
8.
by jmalicki 11d ago
That's not implementing tail calls breaks things, that's bad design of implementing tail calls breaking things. The whole idea of "let's change semantics to make it easier" is dumb. If you want guaranteed tail calls
9.
by jmalicki 12d ago
It's not a change in semantics of compiled code. It is only a change of whether or not the code will compile .
10.
by jmalicki 12d ago
> because it means subtle changes (introducing a destructor, re-ordering code, etc) can change semantics without you realizing it. No, it won't change semantics - if you say @musttail or similar, it will simply fail to compile if yo
11.
by jmalicki 13d ago
> using open weights models AWS and Azure give you the same thing for Claude and ChatGPT, no need to be stuck with open weights. They might sometimes store some of it for other purposes (I don't know the specifics), but it is empha
12.
by jmalicki 13d ago
Or just use Azure, AWS, etc. for Claude/ChatGPT inference, where the AI labs never even get your data in their data centers at all. You pay more for it, but if you care that much, use it.
13.
by jmalicki 14d 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,
14.
by jmalicki 14d 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.
15.
by jmalicki 14d 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
16.
by jmalicki 18d ago
> Do you really want the rust compiler to run at the speed of an llm? That... might actually be an improvement?
17.
by jmalicki 18d 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
18.
by jmalicki 18d 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
19.
by jmalicki 21d 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
20.
by jmalicki 22d 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.
21.
by jmalicki 22d 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?
22.
by jmalicki 23d 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
23.
by jmalicki 23d 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
24.
by jmalicki 26d ago
It's no different than generally using AWS.
25.
by jmalicki 26d ago
The foundation of the businesses training AI models. So don't use them for inference.
26.
by jmalicki 26d 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
27.
by jmalicki 26d 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.
28.
by jmalicki 26d ago
It is a completely reasonable comparison for me as a consumer, since they're the costs and benefits that I'll actually get.
29.
by jmalicki 26d 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
30.
by jmalicki 26d 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.

More ›