thatguysaguy

7 ms·

31.
by thatguysaguy 1y ago
Many of the problems (such as the matmul and packing problems) have existing "scoreboards". It would be quite surprising if so many records had already been broken without it coming to the attention of the people who care about th
32.
by thatguysaguy 1y ago
I mean that is changing the dtype, perhaps that's relevant.
33.
by thatguysaguy 1y ago
Are you claiming that for the open problems they give record-breaking solutions for, there were just answers on the web waiting to be found?
34.
by thatguysaguy 1y ago
Can you elaborate on what the unsolved problem you're referring to is?
35.
by thatguysaguy 1y ago
It's not at the language level, but for python JAX has the notion of a pytree (arbitrarily nested combination of lists, dicts, and tuples), and includes map and reduce operations for those objects. It's very convenient!
36.
by thatguysaguy 1y ago
This is clearly just wishful thinking by the article author. Jeff Bezos would show himself to be a coward by... Not doing something he wasn't planning on doing in the first place?
37.
by thatguysaguy 1y ago
Looks like Meta studied their game theory and the news outlets didn't: https://en.wikipedia.org/wiki/Chainstore_paradox
38.
by thatguysaguy 2y ago
The user love/passion on the JAX team is super high. Interacting with them is whatever the opposite of Kafkaesque is.
39.
by thatguysaguy 2y ago
That article links to a tweet by a guy who is not sam altman starting the trend. It was already viral by the time sam did it.
40.
by thatguysaguy 2y ago
I've read a _lot_ of deep learning papers, and this is extremely atypical. I agree with you that if there were any sort of serious implications then it'd be important to establish proof, but in the case of griping on a forum I thi
41.
by thatguysaguy 2y ago
> To answer this, we'll start by attacking Conway's Game of Life - perhaps the most iconic cellular automata, having captivated researchers for decades > At the heart of this project lies... > his powerful paradigm, pione
42.
by thatguysaguy 2y ago
This writing feels so strongly LLM flavored. It's too bad, since I've really liked Alexander Mordvintsev's other work.
43.
by thatguysaguy 2y ago
It's nice that that the url to an article about studying mysticism with science could also be read as "Frontier Sin"
44.
by thatguysaguy 2y ago
Yeah I think every lab would love to do this and the field has been thinking about it forever. (Search lifelong learning or continual learning on Google Scholar). I don't think a technological solution is likely enough that we should p
45.
by thatguysaguy 2y ago
If you think about it, that's basically what's going on if someone does a quantum mechanics simulation.
46.
by thatguysaguy 2y ago
If you find this appealing, make sure to do it with a medical professional. If you still want to try it by yourself, read Felix Hill's suicide note first.
47.
by thatguysaguy 2y ago
I know a number of otherwise pleasant people who have as a political/philosophical view that it's not rude to insult certain identity groups. I wouldn't say I find it triggering when someone says something about men, just ann
48.
by thatguysaguy 2y ago
Some cold water on the claims here: https://open.substack.com/pub/stoneztao/p/the-new-hyped-gene... tl;dr: The way they're able to claim a massive speedup is by simulating much simpler scenarios than tho
49.
by thatguysaguy 2y ago
> Monitor Apple, Google, Microsoft, and Meta stock. Notify me if any of them start trending toward being undervalued If you actually had a system that could detect when a stock was "undervalued" you should just trade directly o
50.
by thatguysaguy 2y ago
Yeah from the title I thought this was going to be about leetcode problems, but this is truly something that comes up regularly. Where are these dev jobs where _don't_ have to figure out some mysterious issue in a barely maintained Git
51.
by thatguysaguy 2y ago
It's very bad in CS as well. See e.g.: https://arxiv.org/abs/1807.03341 IIRC there was also a paper analyzing how often results in some NLP conference held up when a different random seed or hyperparameters were u
52.
by thatguysaguy 2y ago
I think it's clear that this paper has stood the rest of time over the last 20 years. Our estimates of how much published work fails to replicate or is outright fraudulent have only increased since then.
53.
by thatguysaguy 2y ago
The claim was that Python is only used for prototyping at Google, not that people are writing the AI frameworks themselves in pure Python. JAX is obviously implemented in C++, but the scientists running the training runs which cost millions
54.
by thatguysaguy 2y ago
I have worked on a team which trains foundation models
55.
by thatguysaguy 2y ago
That is quite the claim, any source? Edit: I guess I'm not sure on whether large training runs count as prod or not. They're certainly expensive and mission critical.
56.
by thatguysaguy 2y ago
Not sure on the specific combination, but since everything in Jax is functionally pure it's generally really easy to compose libraries. E.g. I've written code which embedded a flax model inside a haiku model without much effort.
57.
by thatguysaguy 3y ago
Sounds like these guys didn't use custom kernels, but BitNet did.
58.
by thatguysaguy 3y ago
To think of non-evil versions just consider cases where right now there's no voice actor to replace, but you could add a voice. E.g. indie games.
59.
by thatguysaguy 3y ago
I'm 100% going to clone my voice and use it on my discord bot.
60.
by thatguysaguy 3y ago
Completely agree. He ends up sounding kinda amateurish but that's only because (unlike most other podcasters) he's willing to ask questions deep inside the domain of the interviewee. (Amateurish w.r.t. the domain, not as an interv

More ›