JackRumford

6 ms·

1.
by JackRumford 3y ago
Not all other nations - just Palestine. Because Israel is in position of power.
2.
by JackRumford 3y ago
yes but the size of the portfolio is the issue here
3.
by JackRumford 3y ago
I solve exponentially harder problems. When I’m im at the peak, I will start pursuing another field that is foreign to me and, again, grind to the top. A cycle like this takes me about two years depending on the sub topic. I have no interes
4.
by JackRumford 3y ago
Concerning /s
5.
by JackRumford 3y ago
Almost everyone I know in the dev community agree that both generative art and code is fine. Most people don't care about copyright, to say the least. At least the people I know in real life and chat with online. Only thing I've s
6.
DeepMind shows Promptbreeder, an AI that self-improves via recursive prompts(arxiv.org)
2 points by JackRumford 3y ago|0 comments
7.
Understanding LLMs Through the Problem They Are Trained to Solve(arxiv.org)
2 points by JackRumford 3y ago|2 comments
8.
by JackRumford 3y ago
Abstract: The widespread adoption of large language models (LLMs) makes it important to recognize their strengths and limitations. We argue that in order to develop a holistic understanding of these systems we need to consider the problem t
9.
Towards Robust Continual Learning with Bayesian Adaptive Moment Regularization(arxiv.org)
1 points by JackRumford 3y ago|0 comments
10.
by JackRumford 3y ago
Criticisms of Elon Musk are not solely based on ideological differences, but also on concerns regarding his behavior, management style, and the working conditions at his companies, which some believe warrant scrutiny irrespective of his per
11.
by JackRumford 3y ago
I suspect they are trying to compete in the console market without having a console per se. AFAIK the iPhone 15 Pro has 1/3rd of the TFLOPs of the Playstation 5. They also have a smooth TV-connectivity system. Most people who have cons
12.
by JackRumford 3y ago
yep. this guy keeps up to date with quantised models: https://huggingface.co/TheBloke
13.
by JackRumford 3y ago
"With FLASHATTENTION (Dao et al., 2022), there is negligible GPU memory overhead as we increase the sequence length and we observe around 17% speed loss when increasing the sequence length from 4,096 to 16,384 for the 70B model."
14.
by JackRumford 3y ago
HuggingFace model: https://huggingface.co/migtissera/Synthia-7B-v1.3
15.
Synthia-7B-v1.3 hits 64.85 on 4-evals; LLaMA-2-70B-Chat at 66.8(twitter.com)
2 points by JackRumford 3y ago|3 comments
16.
by JackRumford 3y ago
These sites say 154B: https://www.ankursnewsletter.com/p/gpt-4-gpt-3-and-gpt-35-tu... https://blog.wordbot.io/ai-artificial-intelligence/gpt-3-5-t...
17.
by JackRumford 3y ago
Ashamed to say but I do all these things on my main OSs for ~15 years and I never had any problem (IIRC). Yes, I'm not huge on security and I realize what this could do, but I figure the chance is as big as getting it through a channel