fenomas

25 ms·

361.
by fenomas 3y ago
That course is extremely good. Particularly the first six videos start from literally nothing (but basic math/coding), and build all the way up to AI models that were state of the art a matter of years ago. The way Andrej explains thin
362.
by fenomas 3y ago
You're talking about a different usage than GP is. TFA explains the distinction: > Garner tracks the evolution of the third-person-singular “they,” starting in the 1300s, when it sometimes referred to unspecified individuals. (It’s
363.
by fenomas 3y ago
I was suggesting that TFA's author should know that the red-handed stuff doesn't make sense, not that its typical reader should.
364.
by fenomas 3y ago
This article seems weird. Omegaverse is quite well-known as fandoms go[0], so it seems like exactly the kind of thing you'd expect an AI trained largely on web content to know about. And if there's any doubt you could always check
365.
by fenomas 3y ago
I've been doing the same for years and I'm really happy with it. But the one big annoyance is events - I've never found a satisfying way of writing a JSdoc that says what events a class emits, or what parameters the events ha
366.
by fenomas 3y ago
Thanks! I know basically nothing about the Minecraft thing - I didn't hear about it until a week after it launched, and by that time the multiplayer servers were dead and the game was abandoned. Apparently the Babylon.js folks had some
367.
by fenomas 3y ago
Personally I went with HTML for my web game UI and I've not had any reason to regret it. I mean game UIs are thorny and annoying, but it would be a lot worse working inside the canvas. My biggest surprise was, before I jumped in I figu
368.
by fenomas 3y ago
Wow, as someone who's been maintaining a JS voxel engine ("noa") for some years now it's surprising to find a bunch of influences here I hadn't seen before. I also got started from the algorithms on Mikola Lysenko&#
369.
by fenomas 3y ago
Sure, and they sometimes say "Manchester City is lined up on the field." There's no single rule, it's a question of what's emphasized.
370.
by fenomas 3y ago
I don't think this has anything to do with grammatical matters like count/mass nouns. It's just that a word for a collective can often be used either to refer to the collective as a whole, or to its individual members. > M
371.
by fenomas 4y ago
It seems a bit like they're saying: "ohhhhh, did you people think we were an image hosting site? Heavens no, we're an SNS and always have been. We'll update our rules to clarify this". Which seems... aspirational?
372.
by fenomas 4y ago
This strikes me as being at least as big a deal as the NSFW bit. For 10+ years, imgur has been many people's go-to way of hosting an image they want to post to a forum or whatever. Many of those people will have paid little attention t
373.
by fenomas 4y ago
No, I presented it as a counterexample to the claim that LLMs just statistically regurgitate existing data.
374.
by fenomas 4y ago
> it was literally shown the game with only the legal moves It's shown token sequences only. It has no idea they represent a game, or that the game has legal and illegal moves. And more importantly, it has no idea that each token mo
375.
by fenomas 4y ago
> Could the model invent reversi/othello had it not known about it beforehand? You've practically restated the paper's findings! :D The LLM knew nothing about othello; it wasn't shown any rules to be recombined. It wa
376.
by fenomas 4y ago
> LLMs currently statistically regurgitate existing data. This is clearly not true in any meaningful sense - c.f. the Othello paper, examples from the top of this very comment thread, etc. > Can it come up with a corridor when it has
377.
by fenomas 4y ago
Sorry, I don't follow at all. We can assume that every concept humans have conceived of was at some point conceived of for the first time, there's no need to list examples. But how does that relate to the claim here, that purports
378.
by fenomas 4y ago
> Can a machine with access only to statistical patterns in the distribution of text tokens infer the physical structure of reality? We can say, as certain as anything: No. Um. How do you square that claim with the well-known Othello pap
379.
by fenomas 4y ago
Respectfully, that sounds like hand-waving. Claiming to know where concepts do and don't come from just leads to questions like "did the natural numbers exist before we did?", which are centuries old and presumably not resolv
380.
by fenomas 4y ago
I don't think this argument holds water at all. Can we imagine that the people who created Rust were able to do so only because they contributed some magical ingredient from their souls, which didn't exist anywhere in the world&#x
381.
by fenomas 4y ago
I don't know if I agree with you there. In my experience LLMs are excellent at forming their own personalities and making their own independent decisions, particularly in areas like giving preferential treatment to people whose online
382.
by fenomas 4y ago
The underlying issue here is that TFA doesn't describe having a discussion about the complexity. If the author asked about complexity as an open-ended question with no fixed answer, that would be a very reasonable way to hear how the
383.
by fenomas 4y ago
Is that relevant to real-world use? You mention Python in TFA - I'm no expert but I believe it does string interning, and caches the results of string hashes, which would suggest real-world performance should be constant over time in b
384.
by fenomas 4y ago
> suggest to them the intern’s argument why the lookup is O(1) and explain that it’s actually O(n). I think any discussion of complexity for this problem is misguided. I mean, it's fine to talk about a hash lookup being o(n) for pse
385.
by fenomas 4y ago
> I believe that in about fifty years’ time it will be possible to programme computers, with a storage capacity of about 10^9, to make them play the imitation game so well that.. Turing wrote that in 1950, so I suppose his intuition was
386.
by fenomas 4y ago
An attention mechanism is one or more layers in the neural network. When someone talks about attention altering the input vectors, they're referring to what those layers are doing and how data is transformed as it passes through them.
387.
by fenomas 4y ago
A couple of times in recent years Neil Gaiman has criticized museums for talking about the comics Lichtenstein copied but not crediting the artists who drew them. E.g.: https://twitter.com/neilhimself/status/145954
388.
by fenomas 4y ago
Oh that's wild. Actual @substack is/was an open source developer I used to follow, though he seems to have stopped tweeting some years back.
389.
by fenomas 4y ago
> So the logical word -> token(s, p) -> id(s) function definitely has 1 position parameter as well. You're missing that it groups in spaces. The position isn't relevant, but "day" is a different token than &quot
390.
by fenomas 4y ago
Huh, unless the demo is broken it seems to tokenize unicode per-byte, rather than per-character. E.g.: "あ" => [40948] "亜" => [12859, 250] "ｱ" => [171, 121, 109]

More ›