Serving 70B-scale LLMs efficiently on low-resource edge devices [pdf](arxiv.org)

229 points by simonpure 2 years ago | 55 comments

Transformers without normalization(arxiv.org)

33 points by kaycebasques 1 year ago | 3 comments

Bit is all we need: binary normalized neural networks(arxiv.org)

85 points by PaulHoule 1 year ago | 36 comments

CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs(arxiv.org)

96 points by matt_d 5 months ago | 12 comments

Language models are injective and hence invertible(arxiv.org)

194 points by mazsa 1 year ago | 136 comments

Transformers as Support Vector Machines(arxiv.org)

209 points by fofoz 3 years ago | 129 comments

AI models can provably hide arbitrary capabilities(arxiv.org)

252 points by sbulaev 13 days ago | 118 comments

An arXiv paper showing AI models can embed encrypted neural network backdoors that provably hide arbitrary capabilities from elicitation-based safety evaluations.

GitLost: We Tricked GitHub's AI Agent into Leaking Private Repos(noma.security)

487 points by ColinEberhardt 3 months ago | 187 comments

Noma Labs vulnerability research post describing how a crafted GitHub issue could trick GitHub's AI agent into leaking private repository contents.

RWKV: Reinventing RNNs for the Transformer Era(arxiv.org)

318 points by ianbutler 3 years ago | 164 comments

arXiv paper introducing RWKV, a model architecture combining Transformer-style parallel training with efficient RNN inference via linear attention.

ArXiv now offers papers in HTML format(blog.arxiv.org)

1130 points by programd 2 years ago | 296 comments

arXiv blog post announcing that research papers are now available in HTML format to improve accessibility for users with disabilities.

Ask HN: Any nerds out there who've read a lot of research papers?

10 points by shriy-singh 1 month ago | 13 comments

Ask HN thread asking readers which research paper abstracts they found most memorable and why, to inspire writing a better paper intro.

Deep Learning Is Not So Mysterious or Different(arxiv.org)

445 points by wuubuu 1 year ago | 112 comments

arXiv paper by Andrew Gordon Wilson arguing deep learning generalization phenomena like benign overfitting and double descent are explainable by soft inductive biases.