redox99

35 ms·

691.
by redox99 3y ago
> Funny thing, I own a vast number of stocks through various means, either directly or via ETFs, and yet never in my life have I had an opportunity to weigh in on the salaries of anyone in the companies I hold equity in. Because obviousl
692.
Are we on the edge of a “ChatGPT moment” for Autonomous Driving?(mobileye.com)
1 points by redox99 3y ago|0 comments
693.
by redox99 3y ago
Sure, I was establishing a upper and lower bound. That's why I said between 2x and 5x margins.
694.
by redox99 3y ago
> $.05/GB That's just a scam rate by AWS. The true price is 1/100th of that, if that.
695.
by redox99 3y ago
So, it's tor but with less hops?
696.
by redox99 3y ago
> Woah, hold the phone! I hate it already
697.
by redox99 3y ago
Yeah, I'm sorry for the incorrect statement. MidJourney claims 3.3 hours of GPU for $10. If that's 3.3 hours of A100, at a rate of $2/hour that's $6.6 of GPU costs. My original statement was based on MJ feeling very expe
698.
by redox99 3y ago
> Inference costs per user are high, at least much higher than most traditional software costs. This way, the costs are aligned with the usage. Meh, the price if MidJourney is like 2 orders of magnitude higher than their inference costs.
699.
by redox99 3y ago
Here's some data from a guy that trains LLMs using GPT4, and has tried both 0314 and 0613 versions of GPT4 https://old.reddit.com/r/LocalLLaMA/comments/16bi7bs/any_ben... https://twitter.
700.
by redox99 3y ago
Do you actually mean the non determinism with 0 temperature, or just the non-zero temperature of the ChatGPT UI?
701.
by redox99 3y ago
Based on the results I assume it's Llama-2-70b-chat and not Llama-2-70b
702.
by redox99 3y ago
These models aren't even trained on FP32. The original format is FP16. And quantizing to INT8/FP8 is almost lossless. But yes, 2.5 bits per weight is pretty insane.
703.
by redox99 3y ago
That's absolutely false, unless you mean running llamacpp with half the layers on the CPU and half the layers on the GPU. And that was like 2t/s.
704.
by redox99 3y ago
Unrelated. What matters for that is prompt processing time (which is in the high hundreds of tokens per second).
705.
by redox99 3y ago
At least in my country, to diagnose you with high BP they'd have you wear a holter monitor for 24 hours.
706.
by redox99 3y ago
Outside of Macs, llama.cpp running fully on the cpu is more than 10x slower than a GPU.
707.
by redox99 3y ago
Moderation really varies a lot depending both on the language and the group being criticized. For example on twitch, Spanish streamers[1] say a lot of things that would get an English streamer canceled and/or banned for saying it. And
708.
by redox99 3y ago
LLMs are deterministic. However to make them more "creative", the outputs of the LLM can be sampled using something called temperature that adds some randomness. You can set the temperature to 0 and it will be deterministic. Havin
709.
by redox99 3y ago
Yes
710.
by redox99 3y ago
As of right now they don't turn off 3rd party cookies when you enable this, so no. This objectively decreases your privacy.
711.
by redox99 3y ago
> Those aren't topics. Don't worry, they'll keep the first version mild, and in a few years they'll update it with more granularity. > The real thing you should be worried about is third party cookies. They don&#x2
712.
by redox99 3y ago
These LLMs with cosine learning rate are usually decayed just to 1/10th the peak LR. Even if you used exponential decay, your LR on your final step could still be arbitrarily large depending on how fast you configure it to decay.
713.
by redox99 3y ago
Nah, you can check any math or physics book that's many dedcades old and they had this same kind of problems.
714.
by redox99 3y ago
> Wolfram Alpha can solve all of the problems in a typical math textbook I mostly disagree (and I used WolframAlpha thoroughly during my engineering education). Wolfram can solve well encapsulated tasks (solve for X, find the integral, e
715.
by redox99 3y ago
You can't imagine where X, owned by saudis and the world's richest man, got $6 million from?
716.
by redox99 3y ago
Meh, I was just ballparking. But add a threadripper, compatible motherboard, lots of RAM, fast storage, a few beast PSUs, and tax, and you're probably around 18k. So you ain't getting 2 of these machines for 25k. Plus you'll
717.
by redox99 3y ago
25K will either get you money for renting servers, or allow you to jerry rig yourself something like a 8x4090 machine. The servers that most people would like to run (8xA100, 8xH100) are over 200k, and even if you had the money they're
718.
by redox99 3y ago
I want to congratulate you based on your current grant recipients. They are all very impactful individuals/teams. It shows that you clearly did your research and put the money in the right places.
719.
by redox99 3y ago
Aside from all the points mentioned by the other comments, there's also the factor that Zuck probably finds this kind of research cool, and considering he owns the majority of the votes, he can do almost whatever he wants with Meta.
720.
by redox99 3y ago
Prisma is better than any Python ORM in my opinion.

More ›