krackers

33 ms·

781.
by krackers 1y ago
What does "inferring the UI content mean". You said you have some knowledge of how it pulls it off, is it using accessibility API and apple events behind the scenes? Those are the only two APIs I know of on mac that can do this (b
782.
by krackers 1y ago
So snappy, it's almost faster to mac os 8 in a browser than it is to use mac osx.
783.
by krackers 1y ago
Right that's the explanation I've heard too (and I think Karpathy even said it so it's not some fringe theory). I wasn't dismissing the hypothesis but asking out of genuine curiosity, since this feels like something that
784.
by krackers 1y ago
Neat finding, does it generalize to larger samples? Someone should randomly generate a few thousand such strings, feed it to 4o or o3, and get some accuracy results. Then compare the accuracy in cases of counting individual letters in rando
785.
by krackers 1y ago
I see this claim repeated over and over, and while it seems plausible, this should be an easily testable hypothesis right? You don't even need a _large_ model for this because the hypothesis you are testing is whether transformer model
786.
by krackers 1y ago
Yeah this part was confusing, because it's only mentioned halfway through the article that the attention step can only be batched across matching context-window sizes.
787.
by krackers 1y ago
More in https://twitter.com/antimatter15/status/1926459086352142663#...
788.
by krackers 1y ago
I haven't been following image generation models closely, at a high level is this new Flux model still diffusion based, or have they moved to block autoregressive (possibly with diffusion for upscaling) similar to 4o?
789.
by krackers 1y ago
I don't think it's clear that human languages are context sensitive. The only consistent claim I can find is that at one point someone examined Swiss German and found that it's weakly context sensitive. Also empirically human
790.
by krackers 1y ago
You can train LLMs on the output very complex CFGs, and it successfully learns the grammar and hierarchy needed to complete any novel prefix. This is a task much more recursive and difficult than human languages, so there's no reason t
791.
by krackers 1y ago
>is something implicitly present in both our minds and an LLM The LLM doesn't start with any real structure besides the network of ops though. If there is any induced structure, it's learnable from the data. And given enough da
792.
by krackers 1y ago
Restricted to linguistics, LLM's supposed lack of understanding should be a non-sequitur. If the question is whether LLMs have formed a coherent ability to parse human languages, the answer is obviously yes. In fact not just human lang
793.
by krackers 1y ago
Thank you for chiming in! LTO wouldn't really matter for "delinking" exported symbols of shared libraries though, would it? The exported functions must necessarily follow platform ABI convention, and so long as those are copi
794.
by krackers 1y ago
I don't see what ASLR has to do with the inability to convert a shared to static library though. In fact a shared library must be position independent so if anything it would make the job easier. For going from static (non-PIC) to shar
795.
by krackers 1y ago
I'm not sure I follow why 'I flipped them'. My understanding a shared library is closer to an executable than it is to an object file, so it is in that sense that it is "prelinked and ready to map into memory" just
796.
by krackers 1y ago
I think this finally answers a question I've had since I started learning programming. All my life I've been told "you can't convert a shared library to a static library". When you try to probe for more details, all
797.
by krackers 1y ago
Someone [1] inspected dimensions of the embedding component of model and it seems GP was on the right track. Assuming I understood correctly in [2], it does seem to be the embedding of the input tokens which is passed directly into each lay
798.
by krackers 1y ago
Someone extracted out the dims in https://twitter.com/cccntu/status/1925043973170856393#m It seems to be embedding from 262k possible vocab tokens down to 256 dims. 262144 matches the same vocab size used for the
799.
by krackers 1y ago
I am perfectly aware of that. I don't believe other LLMs have such embeddings per layer, only the usual weights, so these per-layer embeddings seem to be distinguished from weights in some way. Afaik trying to play the same "cache
800.
by krackers 1y ago
Thank you, that helped a bit, although it's still not clear what exactly those parameters _are_. "Per-Layer Embedding (PLE) parameters that are used during model execution to create data that enhances the performance of each model
801.
by krackers 1y ago
What is "Per Layer Embeddings"? The only hit I can find for that term is the announcement blogpost. And for that matter, what is >mix’n’match capability in Gemma 3n to dynamically create submodels It seems like mixture-of-exper
802.
by krackers 1y ago
https://news.ycombinator.com/item?id=43076987
803.
by krackers 1y ago
The unreasonable effectiveness of Enumerative Combinatorics
804.
by krackers 1y ago
For how much they talk about "subculture perception" in the announcement, holo genuinely seemed like some sort of counterculture. Compared to the glossy iOS, Holo was dark, gritty, Tron like. Using it signaled that you were a nerd
805.
by krackers 1y ago
* Ask clarifying questions: Get failed for not being able to handle ambiguity * Don't ask clarifying questions: Get failed for making assumptions and not identifying business requirements before implementation
806.
by krackers 1y ago
Current "material design" is the anthesis of what "material design" was originally supposed to be.
807.
by krackers 1y ago
Wasn't the point of REST supposed to be runtime discoverability though? Of course REST in practice just seems to be json-rpc without the easy discoverability which seems to have been bolted on with Swagger or whatnot. But what does MCP
808.
by krackers 1y ago
Turns out that https://www.fortressofdoors.com/four-magic-words/ was right and all you need to do in training is have the LLM meditate on a single example.
809.
by krackers 1y ago
It fits Apple's modus operandi to enforce things UI/UX wise, I assume in this case they don't want end-apps to be able to bypass the compositor (and e.g. prevent alerts from showing on the screen or whatnot). They used to all
810.
by krackers 1y ago
>how to do fullscreen graphics on my Mac You can't, you don't have direct access to the framebuffer. Unless by "fullscreen" you just mean spanning from end-to-end in which case you can create an opengl or metal view a

More ›