redox99

40 ms·

841.
by redox99 3y ago
And Cruise with tens of cameras, lidars, radars, ultrasonics rear ended a bus[1]. I'm not trying to bash Cruise. My point is that shit happens , we should look at the complete statistics, not one offs. [1] https://www.kron4
842.
by redox99 3y ago
> I'm willing to admit that I'm missing something here, but I'd love it if someone could enlighten me. It's basically the same as with Midjourney. Before Midjourney I'd have to spend quite some time organizing wi
843.
by redox99 3y ago
LLMs will just scrape the website like they do with the rest of the internet. For Alphabet, Microsoft and some others they don't even need to do that as they already have everything stored for search results.
844.
by redox99 3y ago
Source?
845.
by redox99 3y ago
Wait till you hear about test and validation set.
846.
by redox99 3y ago
(I'm not an expert on LLMs, this is just my guess) No tokens is like no compression. It's much less efficient. My previous sentence is 12 tokens (24 bytes), or 58 characters (58 bytes). Your self attention block will be able to ha
847.
by redox99 3y ago
Some cryptocurrencies may or may not be securities. Gary Gensler himself has said that Bitcoin is not a security.
848.
by redox99 3y ago
That's kind of expected since GPT2 was already relatively decent at summarizing.
849.
by redox99 3y ago
People that run 7B 4bit is probably because they don't have the memory for more. And there are tests showing that quantizing a bigger model is always better than running a smaller model full precision (i.e LlaMa 13B 4bit > 7B fp16)
850.
by redox99 3y ago
First try on GPT 3.5 (through the chat web) I get > Based on the information provided, Shape A is red and Shape B is the other shape. However, it is not explicitly stated what color Shape B is. Therefore, without further information, we
851.
by redox99 3y ago
WizardLM-30B-Uncensored (4bit) > There are two shapes each a different color, one shape is red and one shape is blue. Shape A is red. Shape B must be what color? > Shape B can only be the other available color which is blue since we k
852.
by redox99 3y ago
You can see the messages if you take a long exposure shot (2s) from your phone Here are the photos from my phone https://imgur.com/a/WOqbQVo
853.
by redox99 3y ago
> USB-C @ 100W is 20 volts at 4.5 amps Clearly one of those numbers is wrong
854.
by redox99 3y ago
None of these models are even close to GPT 3.5, let alone GPT4.
855.
by redox99 3y ago
How? 24GB VRAM can only handle 30B 4bit. Are you offloading some layers to the CPU? That kills performance AFAIK.
856.
by redox99 3y ago
It's trivial to avoid infinite loops by adding some exit condition in case it takes too long, which is probably what you want here. And obviously trivial to avoid recursion too.
857.
by redox99 3y ago
Maybe there are some small tests (GPT4 comparing it to other contenders), but it's mostly other people's experience trying a lot of these models. Also it being 30B makes it already quite stronger than those in the leaderboard. And
858.
by redox99 3y ago
That's obviously not true if you look at the deliveries per quarter chart.
859.
by redox99 3y ago
Unfortunately the Arena is missing some of the strongest "open" models, such as WizardLM Uncensored 30B. In fact they don't have any Llama 30B/65B based models, just 13B models.
860.
by redox99 3y ago
For a Linux user, you can already build such a system yourself quite trivially by [...]
861.
by redox99 3y ago
The animation when you drag a photo is extremely nice
862.
by redox99 3y ago
3090 can handle the ~30B models quantized to 4 bits.
863.
by redox99 3y ago
It's kind of creepy when GPT4 uses such life-like analogies. You're an AI, you've never used crayons or played with legos.
864.
by redox99 3y ago
A100s have a much worse performance per dollar than 3090s/4090s (which are the direct competitors to the 7900x)
865.
by redox99 3y ago
I don't think anybody claims it's unoptimizable. It's just a harder task, compared to a more constrained system.
866.
by redox99 3y ago
Can you provide any evidence of the US forcing someone to update their canary?
867.
by redox99 3y ago
> it’s currently the top OSS chat model that can be used commercially on the chatbot arena score board To be fair, that filters the majority of models in the scoreboard.
868.
by redox99 3y ago
ChatGPT is not a PWA, so it does not work very well if you do that. However I agree that if it were a PWA, it could work very well.
869.
by redox99 3y ago
Could they use a high enough frequency that isn't noticeable? 10kHz maybe?
870.
by redox99 3y ago
Yeah, it's quite clearly wrong when you realize that the batteries would never be able to deliver 50GWh in the first place.

More ›