krackers

57 ms·

901.
by krackers 2y ago
There's a wonderful correspondence between GPU and more conventional SIMD vector terms in the P&H comp arch book. Slide 13 of https://cse.buffalo.edu/~rsridhar/cse490-590/lec/Chapter04.p...
902.
by krackers 2y ago
There's a good derivation of EXP3 algorithm from standard multiplicative weights which is fairly intuitive. The transformation between the two is explained a bit in https://nerva.cs.uni-bonn.de/lib/exe/fetch.p
903.
by krackers 2y ago
I'm glad someone brought Picchi 2 to life https://www.youtube.com/watch?v=WqjUlmkYr2g
904.
by krackers 2y ago
But this still doesn't need AI, unless we're calling OCR "AI" now
905.
by krackers 2y ago
Also seen in Soulver.
906.
by krackers 2y ago
Everything says "killed by saturation". Is there another way to be killed?
907.
by krackers 2y ago
>We are now confident we know how to build AGI as we have traditionally understood it But we don't even have good definitions to work with. Does he mean AGI as in "sentience", AGI as in "superintelligence", AGI a
908.
by krackers 2y ago
I'm not really a bear on llms, but using it for explanations in this way needs a lot of caution. It's very easy to give an argument that "seems" right at first glance, e.g. https://xkcd.com/803/
909.
by krackers 2y ago
Good perspective. Maybe it's because people are primed by sci-fi to treat this as a god-like oracle model. Note that even in the real-world simulations can give wrong results as we don't have perfect information, so we'll pro
910.
by krackers 2y ago
Yeah we know positive reinforcement is better than negative one for humans, why wouldn't you use the same approach with LLMs. Also it's better for your own conscience.
911.
by krackers 2y ago
>that forum post no longer exists and no longer shows up in search results I dream of someone taking the internet archive data, capping it at 2010 or so, then making a search engine out of it. I mean if AI companies are looking to gobble
912.
by krackers 2y ago
The related list of ten Marcus-worthy tasks is probably more worth reading, since it at least provides some falsifiable goalposts. https://news.ycombinator.com/item?id=42560545 I'd argue 1, 2, maybe 6 are effectively a
913.
by krackers 2y ago
>simply because they're probabilistic You can make them deterministic by fixing the temperature. But even still, there is some intrinsic error rate: it's like an oracle machine where some subset of queries will simply give you
914.
by krackers 2y ago
>A surprise in recent years has been the speed of takeover, which is measurable within an hour. These findings lead us to a new hypothesis on the origin of REM sleep If that's true, does it explain the phenomenon of "mental&quo
915.
by krackers 2y ago
>studying their reactions Well what are their reactions?
916.
by krackers 2y ago
Only thing left to do is use ChatGPT to generate random "OpenAI is X" articles.
917.
by krackers 2y ago
That's almost word for word what geohotz said last year?
918.
by krackers 2y ago
"We don’t actually want them to do math for the sake of replacing calculators" - couldn't the article just end here? People aren't giving it multiplication problems or asking it to count letters because they want to know
919.
by krackers 2y ago
Even just using plain C, CoreFoundation and grand central dispatch make for a really nice standard library. The fact that CoreFoundation bridges with Cocoa types is icing on the cake, and just makes every feel so cohesive in a way that&#x27
920.
by krackers 2y ago
Yeah this part is what makes the high performance even more surprising to me. The fact that LLMs are able to do so well on visual tasks (also seen with their ability to draw an image purely using textual output https://simonwilli
921.
by krackers 2y ago
It might be possible if the browser plays some high frequency inaudible tone that's picked up by the phone
922.
by krackers 2y ago
It's in this presentation https://www.youtube.com/watch?v=qbIk7-JPB2c The most significant part I took away is that when safety "alignment" was done the ability plummeted. So that really makes me wonder how m
923.
by krackers 2y ago
>Disregarding that use case, it shows that GPUs can actually do things like fprintf or open sockets. Can you elaborate on this? My mental model of GPU is basically like a huge vector coprocessor. How would things like printf or sockets w
924.
by krackers 2y ago
Only before 10.7 from userspace with CGDisplayBaseAddress https://developer.apple.com/library/archive/documentation/Gr... After 10.7 (and certainly post-Metal) I don't think the framebuffer is accessible
925.
by krackers 2y ago
If you're eating that much, be careful about cadmium and lead.
926.
by krackers 2y ago
You can look at toxoplasmosis as an example, synthesizes dopamine and is somehow able to selectively modulate attraction to predators.
927.
by krackers 2y ago
And all real-world computers are also just finite state machines. But we obviously model computers as turing machines for a reason.
928.
by krackers 2y ago
Double-entry: each row of your DB stores {debit account, credit account, amount}. Single-entry: each row of your DB stores {account, delta}. With double-entry you are guaranteed via your schema that sum(debit delta) = sum(credit delta), tha
929.
by krackers 2y ago
>which has become completely diluted at this point. Isn't it common in TCS to consider the class X with oracle access to decide L though?
930.
by krackers 2y ago
So having individualized, original thoughts not arising from the herd-mind is considered "unusual." What a world we live in.

More ›