fenomas

28 ms·

391.
by fenomas 4y ago
Wouldn't the simpler explanation be that the pretraining teaches the model to extract human voices and phonemes from general audio, and so extending the model to other languages is then easier because the model is already working from
392.
by fenomas 4y ago
Oh neat, I hadn't seen Vue3/script-setup, it looks very similar to Svelte. The magic I referred to was basically just wrapping reactive logic around variable assignments, so if you're familiar with Vue3 then you it sounds l
393.
by fenomas 4y ago
Mine is web-based - a bunch of HTML DOM that lives on top of a canvas rendering the game. Afraid I don't know much about desktop/native, sorry!
394.
by fenomas 4y ago
Yeah, it's very subjective - the above has been my experience, but I wouldn't argue that it will necessarily reflect anyone else's. But for why I go so long without touching the UI code, it's just because the project is
395.
by fenomas 4y ago
Yes, great point - I meant to say that using CSS animations is hairy if one is using Svelte, not that they're hairy generally.
396.
by fenomas 4y ago
Co-opting my comment to add another thought. For me, Svelte's big superpower is that it does a great job of looking like plain JS sitting next to templated HTML, and the two just magically react to each other. The reason this is huge
397.
by fenomas 4y ago
I switched my game's UI from Vue2 to Svelte maybe a year ago, and I agree with almost all of the article. One thing I think the author overlooks: the appeal of Svelte's built-in animations and transitions is that they work correct
398.
by fenomas 4y ago
I think you've made TFA's point better than TFA does. "Talking doesn't work as well as people think" is going to be stuck in my head for a while.
399.
by fenomas 4y ago
Ah thanks - for others here is the link, though TFA may not necessarily have used the same prompt: https://docs.google.com/document/d/1_eg_jiUE5y8e5zeiz5HCGDc2... Based on that it looks like the author asked all 2
400.
by fenomas 4y ago
Am I missing something? Most of TFA is about Bard failing to answer with rhyming words, but in the only prompts shown the author doesn't actually ask for rhyming words. He just says the hint and the name of the puzzle. Is this not si
401.
by fenomas 4y ago
Not -4, but I tested the ASCII gameboard idea a bunch on ChatGPT (3.5) and could not make any headway. It would consistently announce a move and then print out the same gameboard it had been given (i.e. the game state before its move). It s
402.
by fenomas 4y ago
> Or would you say that it's unacceptable to use race as a variable at all, You're arguing against a claim that neither TFA nor anyone here made. TFA doesn't suggest doctors should ignore past research that examined race -
403.
by fenomas 4y ago
> Should the research that tells you which drug to give be banned? No, and nobody suggested anything like that, so if you're trolling please don't. The point of TFA is that a study reporting "drug X has greater risks for b
404.
by fenomas 4y ago
> This is a bit ridiculous. It's well known in medicine that members of different ethnic groups.. Emphasis added. Race != ethnicity; TFA argues against using the former and suggests using the latter (with other things). > Some
405.
by fenomas 4y ago
Huh, thanks. Seems like a weird moat to hide it behind but I guess they know what they're doing..
406.
by fenomas 4y ago
Can Midjourney be used other than via discord bots yet?
407.
by fenomas 4y ago
It breaks the pattern but you might also like: 自演リプ (jien-ripu) Literally it's "reply" together with a term for an actor performing a work that they also wrote. And together the phrase is net slang for sock-puppeting.
408.
by fenomas 4y ago
Here's another interesting one that gamers might recognize, JP-from-CN: > 竜頭蛇尾 (ryuutou dabi) "Dragon head, snake tail" Blizzard's Overwatch team decided that would make a cool family motto for a ninja clan, so it fea
409.
by fenomas 4y ago
Pre-training refers to unsupervised training that's done before a model is fine-tuned. The model still starts out random before it's pre-trained. Here's where the Othello paper's weights are (randomly) initialized: http
410.
by fenomas 4y ago
> what they did here is take the core pre-trained GPT model, did Supervised Fine Tuning with Othello moves They didn't start with an existing model. They trained a small GPT from scratch, so the resulting model had never seen any in
411.
by fenomas 4y ago
> It found a model. That doesn't mean it can use the model. It used the model in the only way that was investigated. The researchers tested whether the AI would invent a (known) model and use it to predict valid moves, and the AI di
412.
by fenomas 4y ago
Sorry, but what does anything you've said there have to do with the Othello paper? The point of that paper was that the AI was given nothing but sequences of move locations, and it nonetheless intuited the "world model" neces
413.
by fenomas 4y ago
TFA claims GPT-2 isn't open. Is it that the trained weights are open but the project that did the training isn't?
414.
by fenomas 4y ago
Question for someone who follows this space: are there any significant models/tech/etc that OpenAI has made open? Either before or after they changed business models. What they've kept private is well-known, but are there a
415.
by fenomas 4y ago
It's legible (ish) due to the limited ways that English characters can group up - e.g. there's only one valid way to split up "writeslikethis". In JP many common words are only 1-2 characters long, so in general even a v
416.
by fenomas 4y ago
Definitely not exactly the same. Kanji and Hanzi are two different character sets - they overlap a lot, but each has common everyday characters that aren't in the other, and sometimes the "same" character is in both sets but
417.
by fenomas 4y ago
I once read an interesting article from a comic book writer arguing that villains are inherently more interesting than heroes - because villains plan, build, and create, and can do so for any of infinite motivations, whereas a hero typicall
418.
by fenomas 4y ago
Fair point ;) Yes, I meant it as a "cow riding a motorcycle" kind of example - concepts the AI knows about in a configuration it hasn't seen.
419.
by fenomas 4y ago
That's what I just said - people are expecting the AI to discern from the prompt whether to answer in fiction mode or fact mode. And my point here is, people who understand the tech should know that only one of those modes actually exi
420.
by fenomas 4y ago
I think this is a big part of it, yeah - I don't remember people calling earlier GPT models dishonest, when the UI presentation was of the AI returning a continuation of the input prompt.

More ›