Philpax

41 ms·

871.
by Philpax 2y ago
https://en.wikipedia.org/wiki/Ward_Cunningham#%22Cunningham'...
872.
by Philpax 2y ago
https://www.interconnects.ai/ has great writing on this; the author is currently working on https://rlhfbook.com/ .
873.
by Philpax 2y ago
https://theonion.com/fbi-uncovers-al-qaeda-plot-to-just-sit-...
874.
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMs(interconnects.ai)
22 points by Philpax 2y ago|0 comments
875.
by Philpax 2y ago
Elon Musk, however, was not.
876.
by Philpax 2y ago
This post is cringe. There's simply no other word to describe it.
877.
by Philpax 2y ago
This [0] is 149 images per second, late ~2023. Not sure what the current SOTA is here on the quality/speed axes, but I'd imagine it's better. [0]: https://www.reddit.com/r/StableDiffusion/comments&#x
878.
by Philpax 2y ago
Photography is theft, especially when taken via point-and-shoot camera. Landscapes need to be painted, portraits need to be sketched etc. Art via photography is 'find and take'.
879.
by Philpax 2y ago
> AI outputs are not created by a conscious agent experiencing affect Who is controlling the AI and driving it to specific outputs? The void?
880.
Scaling the Tülu 3 post-training recipes to surpass the perf of DeepSeek V3(allenai.org)
14 points by Philpax 2y ago|0 comments
881.
No Hype DeepSeek-R1 Reading List(oxen.ai)
4 points by Philpax 2y ago|0 comments
882.
by Philpax 2y ago
See https://www.interconnects.ai/p/why-reasoning-models-will-gen... for more information.
883.
by Philpax 2y ago
If it's fully open source, where's the code for training it? The implementation - at least, theirs - is also not trivial as they've mentioned optimising below the CUDA level to get maximum throughout out of their cluster. I&#
884.
by Philpax 2y ago
Hog is slang for human male genitalia. Given that, the company's name is an imperative command to share photos of such.
885.
by Philpax 2y ago
To some extent, yes. I would not run production off of it, even if it can eek out performance gains on hardware at hand. I'd suggest vLLM or TGI or something similar instead.
886.
by Philpax 2y ago
llama.cpp optimises for hackability, not necessarily maintainability or cleanliness. You can look around the repository to get a feel for what I mean.
887.
by Philpax 2y ago
Yes; this behaviour was designed, it's not an inherent property of LLMs. An infamous example is GPT-4chan, but there are others that demonstrate it's very possible to optimise for anger. Now, agentic anger, that's a more in
888.
by Philpax 2y ago
My personal favourite is PostHog, which sounds like something one might say on Grindr.
889.
by Philpax 2y ago
Of all the problems that I have with Spotify - its frequently changing selection of music, its hostility to power users, its shuffle algorithm, its artist hostile rates, its frequently changing-for-the-worse interface - I can't really
890.
Why reasoning models will generalize(interconnects.ai)
3 points by Philpax 2y ago|0 comments
891.
by Philpax 2y ago
Er, how would that reduce the cost? You still need to train the model, which is the expensive bit. Also, the base model for V3 and the only-RL-tuned R1-Zero are available, and they behave like base models, which seems unlikely if they used
892.
by Philpax 2y ago
We are building machines that can do work and create content, and they are getting better every three months. This year's focus will be on making them work on tasks entirely autonomously (e.g. "agentic"). It has already impac
893.
by Philpax 2y ago
They don't have permission to open-source that code: https://news.ycombinator.com/item?id=42845102
894.
by Philpax 2y ago
People really like their Pebble watches, and this will help keep them alive into the future. In some future, we may see new hardware produced that could use some derivative of the resulting operating system.
895.
Qwen2.5-VL(qwenlm.github.io)
8 points by Philpax 2y ago|0 comments
896.
by Philpax 2y ago
This is blogspam of https://github.com/Jiayi-Pan/TinyZero and https://nitter.lucabased.xyz/jiayi_pirate/status/18828393705... . This also doesn't mention that it's for one specific d
897.
DeepSeek Panic at the App Store(thezvi.substack.com)
36 points by Philpax 2y ago|3 comments
898.
Mixture-of-Experts (MoE) LLMs(cameronrwolfe.substack.com)
1 points by Philpax 2y ago|0 comments
899.
by Philpax 2y ago
I think what you're saying is consistent with the quote: human curation of SFT data is indeed not complex. There might be extra work on top of that RL, but it's the same work that's been done throughout LLM development.
900.
by Philpax 2y ago
Not really; R1 is post-training on top of V3, which is considerably cheaper than training V3 itself. You can see this in the existence of multiple reproductions of the RL training technique by much smaller labs: https://hkust-nlp

More ›