anon291

7 ms·

1.
by anon291 3d ago
amd consumer gpus don't support rocm. That is basically the entire reason for Nvidia dominance. All Nvidia cards support cuda even the crappy ones
2.
by anon291 3d ago
It's impossible to have an open standard. Hardware accelerators are nothing alike and have different perf characteristics. Each kernel is tuned to the hardware. The idea of writing a performant kernel in opencl is a fantasy. Source: wo
3.
by anon291 4d ago
That's literally not useful for the sort of applications people want to use transformers for.
4.
by anon291 4d ago
As much as I love all these technologies, they don't have any direction that approaches the efficacy of a transformer
5.
by anon291 4d ago
The answer to all ai worries is quite simple. Ai is a technology. While 'agents'may be deployed, the law simply needs to be clear as to who the human agent is who will be held responsible criminally.. Suddenly every problem solves
6.
by anon291 4d ago
Then he should resign
7.
by anon291 4d ago
Forget capitalism. This is a call to end the basic right of people to multiply matrices
8.
by anon291 8d ago
I specifically stick to monasticism because Christianity is the odd one out of all the abrahamic religions in advocating for monasticism, and monasticism started in Egypt which had significant enough presence of Buddhists. As far as I am aw
9.
by anon291 8d ago
Christian and Buddhist monastic traditions are so uncannily similar it's difficult to really think they developed without influence. That's my opinion. Even the Europeans who first set foot in Asia during the age of discovery thou
10.
by anon291 8d ago
I personally think thinking is basically variable but rate precision. If you are in a 4bit mode but need 2x as many tokens you're just doing fp8 with hoops( of course 4bit multiply is faster)
11.
by anon291 9d ago
> while you yourself do the same about modern composers. I literally gave several examples of modern genres in which classical composers still produce good work. I just said they're unlikely to be found in academia
12.
by anon291 9d ago
Pop music is also mostly trash too, but no one pretends otherwise. At least they have the decency to use auto tune.
13.
by anon291 9d ago
> I know people who only listen to a single narrow genre or decade of pop music. How are they different? No one mistakes a pop artist or afficianado for some great connoisseur of culture.
14.
by anon291 10d ago
Modern 'classical' music is a myth. In their day, classical composers were essentially pop performers. Women used to throw their undergarments at mozart... Today classical music is filled with people who want to pretend to be supe
15.
by anon291 11d ago
Yeah as someone whose family left this, you guys are crazy
16.
by anon291 13d ago
I mean at every layer in a transformer, the attention mechanism does a massive state transfer between tokens. In a recurrent transformer, instead of projecting from the latent space to token space after a fixed depth, you take the latent em
17.
by anon291 13d ago
The tokens inside the transformer are only projected into token space to train them. In reality they ought to be treated as their own thing. What's really gone on is you've trained the final projection to be sensible rather than t
18.
by anon291 13d ago
The hidden states of the tokens likely contain more semantic information than can be extracted by the final projection into token space.
19.
by anon291 13d ago
I do the same thing in my head. There is no underlying logic to an llm. Logic is an external construct alien to human like forms of reasoning.
20.
by anon291 13d ago
Why? I talk in my head and then enunciate only that which is relevant. My speech rate inside is incredibly fast.
21.
by anon291 14d ago
Should have said it was post preschool psychosis
22.
by anon291 14d ago
Same model. Same hardware... Depends. If you sacrifice speed then no. Ieee754 is pretty specifically specified, the issue is that it's not associative. If you get the associativity correct, then there's no issue. Associativity usu
23.
by anon291 14d ago
Llms are absolutely deterministic if you want them to be. Still a terrible idea for config
24.
by anon291 14d ago
Whichever you want. There's a million text editors today because people have different approaches. I'm not criticizing having different approaches. I'm criticizing people who start writing their own text editor instead of jus
25.
by anon291 14d ago
The interface of text editors has not substantially changed since the time of interactive editors like VIM and emacs. Line editing is an entirely different paradigm. If you are actually going to change the paradigm with which we edit text a
26.
by anon291 14d ago
Exactly, this my point. If you want to make a great editor, the proper thing to do is start from a good base. Start from emacs and make something like doom emacs or spacemacs, etc. Or start from vim or whatever. Text editing at speed is a s
27.
by anon291 14d ago
It's just easy to install on Windows. Base emacs does exactly what you want it to do. And my emacs v vscode comparison was specifically because of infinite configurability and expandability. VSCode has the same thing with its myriad ja
28.
by anon291 15d ago
My computer stays the same speed because I use software that just does the thing and not anything else. Native toolkits, single purpose, text based, etc. None of this is an issue if you just don't use the crap they push on you.
29.
by anon291 15d ago
I am being pushed to use vs code right now by my team, but we already have a fully programmable and scriptable editor called emacs that is 100000x better. I don't understand why everyone just switches to these random tools. Text editin
30.
by anon291 15d ago
There's no problem with American corporations and a lot of problems with American institutions. That's because corporations are controllable. The decline in governance is not.

More ›