krackers

68 ms·

751.
Sparsity Is Cool(tilderesearch.com)
14 points by krackers 1y ago|0 comments
752.
by krackers 1y ago
Related: https://news.ycombinator.com/item?id=44212855 and https://news.ycombinator.com/item?id=44055347
753.
by krackers 1y ago
Not just hardware, he ruined UI design as well. Everything iOS 7+ (which hasleaked into macOS) is basically his doing, or at least had to be approved by him. Trading Scott Forstall for Johnny Ive was probably the the biggest blunder that Co
754.
by krackers 1y ago
One question, OP, how does cost for this work? Do you pay the LLM inference cost (quite literally if using an external API) every time you run a line that involves natural language computation? E.g. what happens if you call a "symbolic
755.
by krackers 1y ago
Some of this seems a bit related to Wolfram Mathematica's natural language capabilities. https://reference.wolfram.com/language/guide/FreeFormAndExte... It can (in theory) do very similar things, where natura
756.
by krackers 1y ago
There's several ways to arrive at the same bound besides the explicit birthday paradox calculation: * nC2 pairs of objects, each of which has a 1/k chance of matching. Since expectation is linear and it's also known that E[x]
757.
by krackers 1y ago
hasn't this always been the case? I thought one of the whole reasons for the weird design of OpenGL API was that it was a client-server sort of thing to allow for network transparency
758.
by krackers 1y ago
I think you can do this on Android (or at least you used to be able to with Tasker when rooted, since it has [used to have] the ability to read notifications and dismiss notifications)
759.
by krackers 1y ago
GPT-2 can successfully learn to do multiplication using the standard tokenizer though, using "Implicit CoT with Stepwise Internalization". https://twitter.com/yuntiandeng/status/1836114401213989366 If an
760.
by krackers 1y ago
Until I see evidence that an LLM trained at e.g. the character level _CAN_ successfully "count Rs" then I don't trust this explanation over any other hypothesis. I am not familiar with the literature so I don't know if t
761.
by krackers 1y ago
But that only seems to make the situation worse: for all their hand-wringing about "AI safety", by their own benchmark their models seem to do no better than competitors. They don't even have any basis to claim that open-sour
762.
by krackers 1y ago
How many more similar pieces is Anthropic going to put out? Every other weeks it seems like they publish something along the lines of "The AI apocalypse is soon! We created a narrative teeing up an obviously fictional hollywood drama s
763.
by krackers 1y ago
I thought this was about physical libraries as well. Maybe the link is librarians, supposedly if you didn't even know where to begin searching a trained librarian would have been a good person to ask.
764.
by krackers 1y ago
I like the design, very "Delicious Era". That rosetta icon gave me hope it worked on PowerPC though (Rosetta 2 icon is different)
765.
by krackers 1y ago
This seems like a terrible comparison since Google Translate is completely beat by DeepL, let alone LLMs. (Google Translate almost surely doesn't use an LLM, or at least not a _large_ one given its speed)
766.
by krackers 1y ago
If anything you'd think that the neurosymbolic people would be pleased that the LLMs do in fact reason by learning circuits representing boolean logic and truth tables. In a way they were right, it's just that starting with logic
767.
by krackers 1y ago
Yeah this was my thought after skimming the paper as well. The core of the paper seems to be a way to use LLMs' latent knowledge to create a labeled dataset that can then be used for fine-tuning. They even acknowledge that this can onl
768.
by krackers 1y ago
Hasn't the symbolic vs statistical split in AI existed for a long time? With things like Cyc growing out of the former. I'm not too familiar with linguistics but maybe this extends there too, since I think Chomsky was heavy on for
769.
by krackers 1y ago
I do remember that! I don't remember anyone ever coming up with a good explanation for it though, and either my google-fu is weak or all the talk about it has link rotted. I found one comment on HN from 2015 mentioning it though https
770.
by krackers 1y ago
So you can pretend you are an alien learning about human culture.
771.
by krackers 1y ago
I don't get those benefits: randomly generated password is by definition only going to be usable at the site it was generated for. I'm not sure what it means for a password to "encode my identity", but if it includes dev
772.
by krackers 1y ago
>The robot doesn't just need a prediction about the next frame But the residual stream of LLMs doesn't "just" encode the next token prediction, it is high-level enough to encode predictions for a few tokens out, as se
773.
by krackers 1y ago
I'm only a layman but at a high level how does the encoder + predictor of JEPA differ from an LLM? An LLM takes in input, transforms it into an embedding, and makes predictions off that embedding. The only high-level difference I can s
774.
by krackers 1y ago
To me 2011 has slightly more key travel than 2015, but 2015 and M1+ seem to have about same keytravel but M1 is a lot "crisper". M1+ keyboard seems to uniformly actuate wherever you press it, whereas the shallower keycaps of 2015
775.
by krackers 1y ago
Interesting, I never made the connection between dashboard widgets UI and early iPhone UI. It does make sense, early iPhone had a UI that was glossier and more colorful than "metallic" aqua.
776.
by krackers 1y ago
Because most people colloquially know it as "Godbolt".
777.
by krackers 1y ago
Is apple putting out these papers just to justify their seeming inability to properly integrate them into their software?
778.
by krackers 1y ago
Dupe of today's https://news.ycombinator.com/item?id=44212861 ?
779.
by krackers 1y ago
Could it be construed as a dumb question because the details depend on the specific engine you're using? If you have TABLESAMPLE, then the question is trivial. The question doesn't really test anything general other than "are
780.
by krackers 1y ago
They must be doing version numbers based on marketing. From OS X, to macOS 11...15, and now macOS 26.

More ›