cypress66

8 ms·

31.
by cypress66 3y ago
Glyphosate is more of a problem for people that live and work near it. The food on the shelves contains trace amounts of it (parts per billion).
32.
by cypress66 3y ago
Probably gpt4
33.
by cypress66 3y ago
I kinda like Meta, but alphabet is awful (and I almost never hear it)
34.
by cypress66 3y ago
RAG is not always required. If you can fit a whole document/story/etc in the context length, it often performs better, especially for complex questions (not just basic retrieval).
35.
by cypress66 3y ago
> Simple code executed in a thread pool or whatever is so much easier to reason about. reply Hard disagree. It's much easier to reason about async await because you don't need to worry about preemption. You (generally) don&#x27
36.
by cypress66 3y ago
Not really. Most of it can be cached. And prompt processing is quite fast anyway. See vllm for an open source implementation that has most optimizations needed to serve many users.
37.
by cypress66 3y ago
It had the right amount of effort that the parent question deserved.
38.
by cypress66 3y ago
>I'm running gpt2-xl (1.5B params) locally with KV caching at 120ms/token (vs. 450ms without caching). That seems very slow compared to llama cpp?
39.
by cypress66 3y ago
Use axolotl
40.
by cypress66 3y ago
First link is about fp32 which is irrelevant (and 10tflops is a joke). A100 has 300tflops of bf16 Second link is about inference not training
41.
by cypress66 3y ago
1) they don't seem very crazy, gpt4 should mostly handle this 2) they're probably finetuning the model a bit with these instructions
42.
by cypress66 3y ago
> Can't those LLM/text-to-image model rules be embedded in the training/alighnment process instead of being injected before user input? Absolutely. The model would fairly easily learn these rules with enough training even
43.
by cypress66 3y ago
Because enshittification has a specific meaning. It's really about online services that start being "too good", possibly operating at a loss to gain users. And after they have enough users that are "locked in", they
44.
by cypress66 3y ago
Enshittification is subscriptions, software locked features, telemetry and always online systems, and so on. Using these casts, regardless of how good or bad they are, is not enshittification.
45.
by cypress66 3y ago
> 123,000 hotel rooms That sounds kinda low for NYC.
46.
by cypress66 3y ago
No, recommending bitlocker back then was basically a joke to anyone at that time, an obvious way of telling you something wrong. The obvious recommendation back then was LUKS. And even if there were no alternatives they're not going to
47.
by cypress66 3y ago
> I always have issues with LLMs completely forgetting where things are in a scene, or even what parts a given animal has, e.g. saying "hands" when the subject is a quadruped. Sounds like you're using too small of a model.
48.
by cypress66 3y ago
I assume you mean major version? Because the kernel gets updated all the time otherwise.
49.
by cypress66 3y ago
Only few projects (like uniswap) are truly decentralized. Most projects are not, and only claim they'll eventually become decentralized once "they finish development".
50.
by cypress66 3y ago
You seem to be relatively decent at chess, yet you haven't played a human in 40 years once? Why? That's very surprising.
51.
by cypress66 3y ago
Funny how even though I studied EE, it didn't cross my mind it could be an electrical transformer.
52.
by cypress66 3y ago
You don't need "big pharma" for this. Researchers at universities also do these kinds of things because it helps them advance their careers.
53.
by cypress66 3y ago
That was Bing. Chatgpt was always this short. If you're going to significantly finetune the model, you don't need the prompt to be complicated and detailed. Even a single token to let it know "you're in assistant mode no
54.
by cypress66 3y ago
Its from January 2022
55.
by cypress66 3y ago
There's literally nothing about "turbo" as the title mentions either, since it didn't exist yet
56.
by cypress66 3y ago
There's a lot of very low volume niche Chinese phones for reasonable prices. So it doesn't need to be so expensive. You could argue that an iPhone would cost over a billion because you need to develop iOS. But why would you do tha
57.
by cypress66 3y ago
No you don't need to format the %d. The same way you collapsed the loop into the constant 5, you collapse that printf into puts("5\n")
58.
by cypress66 3y ago
I'm replying to the fear that it will leak secret info
59.
by cypress66 3y ago
Use local models if you don't want to send your data to OpenAI
60.
by cypress66 3y ago
Printf is horribly slow compared to puts

More ›