astrange

33 ms·

871.
by astrange 1y ago
I learned there really is a mermaid/merman/merperson emoji and now I just want to know why.
872.
by astrange 1y ago
> Presumably, since we don't observe people saying "near but actually totally incorrect" words in practice https://en.wikipedia.org/wiki/Paraphasia#Verbal_paraphasia > do you ever feel like there i
873.
by astrange 1y ago
I think the em-dashes are OpenAI's way of watermarking their outputs. Makes it easy to spot spambots on Twitter. Well, that and they all have noticeable favorite words.
874.
by astrange 1y ago
That's what reasoning models are for. You can get most of the benefit by saying an answer once in the reasoning section, because then it can read over it when it outputs it again in the answer section. It could also have a "delete
875.
by astrange 1y ago
Those are not two different things. A transistor is a probability function but we do pretty well pretending it's discrete.
876.
by astrange 1y ago
If by "parse" you mean "match", the answer is yes because you can express a context-free language in PCRE. If you mean "parse" then it's probably annoying, as all parser generators are, because they'r
877.
by astrange 1y ago
Who exactly are you imagining is being bribed here?
878.
by astrange 1y ago
Well yeah now it is, otherwise it wouldn't exist. I don't think showing the entire book would be though.
879.
by astrange 1y ago
> The court held that framing and hyperlinking of original images for use in an image search engine constituted a fair use of Perfect 10's images because the use was highly transformative ?
880.
by astrange 1y ago
Google Books is not transformative. It shows you all the same data for the same purpose as they were published for. A better example is Google Image Search. Thumbnails are transformative because they have a different purpose and aren't
881.
by astrange 1y ago
> The tokenizer covers the entire dataset. Well, this is only trivially true. You can feed binary data to the LLM and it probably has tokens that only cover single bytes of that.
882.
by astrange 1y ago
There's nothing new about being able to copyright something that's a transformation of another work. And they definitely aren't exclusively trained on public data.
883.
by astrange 1y ago
"only" and "repeating" aren't accurate here. There's a lot of steps between the pretraining tokens and the LLM. I mean, you can pretty much do whatever you want in the process of making one or running it. For i
884.
by astrange 1y ago
If you think about it, it's not really mechanical if it has an FPGA. It should be made of gears or something.
885.
by astrange 1y ago
> But according to the repo, this project also uses both slice and frame multi-threading (as does ffmpeg, with all the tradeoffs). Oh, I missed that since it doesn't have a separate file. In that case they're likely very simila
886.
by astrange 1y ago
Are the models capable of judging a good SVG? They can't read ASCII art.
887.
by astrange 1y ago
ARMv8.7 added some new instructions for int min/max to replace cmp+csel. (I'm surprised it took them so long to add popcnt.) https://www.corsix.org/content/arm-cssc
888.
by astrange 1y ago
There's no need to do any of that, a table of function pointers to DSP functions works fine. It can be useful to duplicate the entire code for 8-bit vs 10-bit pixels because that does affect nearly everything.
889.
by astrange 1y ago
I don't see why this would support Linux arm64 but not macOS. Anyway, you can just use libavcodec, which is faster (because of frame based multithreading) and doesn't operate on the mistaken belief that it's a good idea to us
890.
by astrange 1y ago
Insurance is not free money and you can't simply distribute risks among your tenants because the risks are correlated. Like I said, you can tell it doesn't work because these businesses don't exist. There are essentially no l
891.
by astrange 1y ago
If you get more space, you need a really good reason not to use that space on more cache. Also, the size numbers are lies and aren't the actual size of anything.
892.
by astrange 1y ago
It's not that different. It has more cores but they're the same cores.
893.
by astrange 1y ago
https://en.wikipedia.org/wiki/Vector_quantization https://sound.stackexchange.com/questions/45719/perceptual-n...
894.
by astrange 1y ago
I mean, that would work. It's going to be the reason AI video compression works; people are okay with an AI model being 1GB but they wouldn't put up with libavcodec being 1GB.
895.
by astrange 1y ago
Until you have to replace a roof, or a tenant destroys the house, or it just doesn't rent for a while and nobody notices a water pipe breaking. It's risky to own a lot of buildings, and worse the risks are correlated if they'
896.
by astrange 1y ago
I live in the most expensive housing market in the world. That's because the value of the land under the houses is so high; the house itself is nothing special. But even then, it's mostly because of Prop 13, and it only works out
897.
by astrange 1y ago
> - Intel Itanium, which (besides some other problems) needed a super-smart compiler (which never existed) so that programs could make use of its potential Well, no such thing is possible. Memory access and branch prediction patterns are
898.
by astrange 1y ago
Homes are depreciating assets. You can't get rich by "buying up homes and doing nothing" because you'd lose money. Nobody is doing this, although a bunch of confused people on social media believe BlackRock is doing it f
899.
by astrange 1y ago
Luckily it doesn't actually copy the style at all. No matter what text you put in the prompt you'll get /something/. Just because you put "studio ghibli anime" in the prompt doesn't mean you're going
900.
by astrange 1y ago
Also, he was calling them ableist because they said crawling was creepy but it reminded him of a disabled man he knew. Though… I'm always surprised how respectful Westerners are about Miyazaki. Meanwhile you read other Japanese directo

More ›