vrighter

12 ms·

241.
by vrighter 5mo ago
reduced by a negative nomber maybe?
242.
by vrighter 5mo ago
the llm is the trained part, the rest is the handwritten part. The sampler is handwritten, not learned.
243.
by vrighter 5mo ago
if a compiler bug is found, it can be fixed. You can't fix an llm.
244.
by vrighter 5mo ago
You need to be able to define exactly what it's abstracting. ex: std::shared_ptr is abstracting over raw pointers, and does refcounting. It is abstracting something but you can actually know exactly what that thing is. An LLM is an a
245.
by vrighter 5mo ago
but you either exchange money and get the food you want, always. Or it's out of stock, so you don't get the food but you keep your money, guaranteed. It's not a good analogy. With an LLM you might ask for a pea, be parted wit
246.
by vrighter 5mo ago
well just run all inference on the cpu, single threaded /s
247.
by vrighter 5mo ago
but even if it didn't it still provided a binary that is mathematically proven (assuming no compiler bugs, which if found are fully fixable, unlike LLMs) to correspond to the code you wrote.
248.
by vrighter 5mo ago
when you do have the option of determinism, but intentionally eschew it in favour of a strictly inferior nondeterministic tool, then yes, it is kinda dumb.
249.
by vrighter 5mo ago
but we already know it's not. self training leads to model collapse
250.
by vrighter 5mo ago
so, half an hour's worth?
251.
by vrighter 6mo ago
Or maybe "how hard is it to find a file you saved 1 minute ago?"
252.
by vrighter 6mo ago
why dmes spacex need an ide?
253.
by vrighter 6mo ago
WDYM mostly? Did you scrutinize the tests even harder than the code?
254.
by vrighter 6mo ago
and it doesn't support NTLM auth afaik! Which is an extra bonus, as I don't remember there being a way to disable it in the windows version
255.
by vrighter 6mo ago
Are the tests testing?
256.
by vrighter 6mo ago
crypto mining firms have specialized asics for mining. They are incapable of doing anything else.
257.
by vrighter 6mo ago
I have never had it automate any tedium for me. Because if I didn't write it, I need to scrutinize all of it, but without the benefit of a prebuilt mental model. Reading through tons of similar looking code (because it's supposed
258.
by vrighter 6mo ago
what i really hate is the "remind me later" buttons. I want to say a plain "no" but the app won't let me. It promises not to respect my decision right there in the popup itself!
259.
by vrighter 6mo ago
GPUs in your average home PC has a longer lifespan. Datacenters run them at full load for very long periods of time. Some datacenters literally burn through hundreds of GPUs a day.
260.
by vrighter 6mo ago
What will happen is that new buzzwords will be invented, and a new fad will take its place. And we will be stuck with the short end of the stick again. You can hope, but shit doesn't really get cheaper for us common folk, ever. :/
261.
by vrighter 6mo ago
I hadn't read that one. Thank you for bringing it to my attention. I loved it
262.
by vrighter 6mo ago
most of today's problem's in this field is because upper management got swindled into thinking that the process doesn't matter, as long as something comes out the other end. Doesn't need to work properly. But this shit
263.
by vrighter 6mo ago
You can ignore precedence in the grammar, and then use a pratt parser or shunting yard or something to parse the precedence. But yes, it does need it, usually. And it's not a huge thing to implement. I usually implement it in the gramm
264.
by vrighter 6mo ago
don't offend blender by comparing it to ai slop.
265.
by vrighter 6mo ago
i still do this today!
266.
by vrighter 6mo ago
except that if you do, you've just implemented a different model, with no way to tell which part of it is wrong
267.
by vrighter 6mo ago
You would need to somehow analyze the prompt, figure out that the user is asking for an addition of two numbers, and selectively enable that filter. If that filter was left enabled permanently then you'd just functionally have a calcul
268.
by vrighter 6mo ago
Funny how much stuff AI can democratize whose only prerequisite is a willingness to spend effort learning to use them, for free. But now "needing" to pay actual money, for an AI agent to write code for you, preventing you from act
269.
by vrighter 6mo ago
how do you disallow it from generating specific things? My point is that you can't. And again, how do you stop it generating certain tokens, but only in certain contexts?
270.
by vrighter 6mo ago
You just said an oxymoron right there. If you're syntax checking every token, you're doing it AFTER the llm has spat out its output. You didn't actually do anything to force the llm to produce correct code. You just reject

More ›