xkapastel

5 ms·

1.
by xkapastel 3y ago
The point is wrong, they don't have any claim to having created deep learning and so their perspective can't be used to argue for the reality of "superintelligence". Even if they did create it, the question of whether &q
2.
by xkapastel 3y ago
This is a strange way to count, the three Turing award winners are not the ones that "created the tech". Many thousands of people did that.
3.
by xkapastel 3y ago
I don't think you can do this, it will just make things up. Language models don't have this type of reflection. Google would need to indicate this out of band, like on the page itself, in order for you to be confident about what m
4.
by xkapastel 3y ago
Using 8-bit still runs out of RAM for both the 3B and 7B models. It's unclear if it's because it still uses more than the available RAM, or if it's just quietly not using 8-bit since it's not implemented.
5.
by xkapastel 4y ago
Amazon did release a 20B transformer called Alexa last year: https://www.amazon.science/blog/20b-parameter-alexa-model-se...
6.
by xkapastel 4y ago
bing cannot do anything besides read some of microsoft's scrape of the web. it can't make a GET or POST request, it can't click buttons, etc
7.
by xkapastel 4y ago
CLIP Interrogator uses BLIP, an image captioning model, as well as trying a bunch of prompts with CLIP. I guess you mean that this model uses the captioning model to generate the complete prompt? Is the code for this one available?
8.
by xkapastel 4y ago
Okay, and that funding and top tier talent was attracted to Google, right? I'm not sure what you meant by that.
9.
by xkapastel 4y ago
Google did not "threaten their own business" by publishing basic research in machine learning. Products threaten businesses, not research.
10.
by xkapastel 4y ago
OpenAI made GPT, not Google. Did you mean "Attention is all you need", the paper that introduced transformers?
11.
by xkapastel 4y ago
did you tell chatgpt to act like jar jar binks or what?
12.
by xkapastel 4y ago
transformer-xl is a totally different net. the thing to keep in mind is that gpt will play along with anything unless they explicitly prevent it
13.
by xkapastel 4y ago
BERT is a simple model that is not capable of answering questions in this manner. For very simple things it might help with that answer box at the top, but that's not what I meant.
14.
by xkapastel 4y ago
It's very different because it will answer the question you asked, rather than answering a question that matches a substring of the question you asked like Google will.
15.
by xkapastel 4y ago
Why do you think this? I don't think there's any reason it would be able to reproduce its own code. It's never seen it so it's not in the weights, and it doesn't have that type of reflection so it can't look it
16.
by xkapastel 4y ago
I don't see how it contradicts the open source definition at https://opensource.org/osd , could you point it out for me?
17.
by xkapastel 4y ago
Why do you think it is not open source? The model weights, model architecture, and dataset are all available.
18.
by xkapastel 4y ago
I linked to the same file you did, that is the "script" I was referring to. And I said that I didn't use it. My point is that the Python API is more interesting than the txt2img script, and it doesn't add any watermark
19.
by xkapastel 4y ago
I wouldn't really say Stable Diffusion marks images as AI-generated. There's a script in the Stable Diffusion repository that will do that, but it's not connected to the model itself in a meaningful way. I use Stable Diffusio
20.
by xkapastel 4y ago
It wasn't unlimited though, there was always a quota. It just wasn't visible.
21.
by xkapastel 4y ago
Puppeteer is headless Chrome for Node.js. https://github.com/puppeteer/puppeteer
22.
by xkapastel 5y ago
thanks for the info! never knew about this, i wonder why it's not on the pico-8 landing page.
23.
by xkapastel 5y ago
no, pico-8 does not have any cpu limits, but it has a code size ("cartridge" size limit) of 32k of lua. edit: incorrect, see reply
24.
by xkapastel 5y ago
Can someone change the title? This is GPT-NeoX-20B, not GPT-3.
25.
by xkapastel 5y ago
The media file is still abundant! What's scarce is the signature.
26.
by xkapastel 5y ago
I don't see how it is correct. If you publish to npm, of course you do it the way Node does it, but that's not very interesting. If you publish a module for consumption by Deno, you're publishing TS files, not JS files, and t
27.
by xkapastel 5y ago
What do you mean? You can use TypeScript or JavaScript, so .ts or .js extensions. I think what you are missing is that, if you use TypeScript in Deno, the extension is .ts. What's confusing is the way Node.js+TypeScript does it, which
28.
by xkapastel 5y ago
That's clearly untrue, there are plenty of artists being paid for art with NFTs. Just go on NFT Twitter and look.
29.
by xkapastel 5y ago
A model of GPT-3's scale is not going to be trained or run on a laptop. OpenAI's restrictions are significant because not many people can run a model that large.
30.
by xkapastel 5y ago
Because it isn't true. There's some money laundering, sure, just like there is in anything that involves money. But if you actually spend time in NFT communities it's not hard to find real people who are spending money either

More ›