Top Links | ML News

ML Newsnew | past | best | rss | submit

1.	DeepSeek Elastic Compute (DSec) (arxiv.org)
	323 points by shenli3514 7 days ago \| hide \| past \| pdf \| 116 comments
2.	Breaking the 1.58-bit Barrier for Ternary LLMs (arxiv.org)
	245 points by matt_d 17 days ago \| hide \| past \| pdf \| 41 comments
3.	An empirical study of harness design for coding agents (arxiv.org)
	225 points by wek 15 days ago \| hide \| past \| pdf \| 59 comments
4.	Dream-RSI: Recursive Self-Improvement through Evolving Worlds (arxiv.org)
	213 points by bananaflag 17 days ago \| hide \| past \| pdf \| 53 comments
5.	Context Language Models (arxiv.org)
	175 points by emersonmacro 2 days ago \| hide \| past \| pdf \| 51 comments
6.	Intelligence per Watt: Measuring Intelligence Efficiency of Local AI (arxiv.org)
	169 points by pythonic_hell 19 days ago \| hide \| past \| pdf \| 65 comments
7.	Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data (arxiv.org)
	158 points by Betelbuddy 16 days ago \| hide \| past \| pdf \| 43 comments
8.	Harnessing the Universal Geometry of Embeddings (arxiv.org)
	122 points by ur-whale 27 days ago \| hide \| past \| pdf \| 46 comments
9.	Cache-to-Cache: Direct Semantic Communication Between LLMs (2025) (arxiv.org)
	109 points by rochansinha 15 days ago \| hide \| past \| pdf \| 22 comments
10.	"As a Language Model": Chat Template Switches LLM Self-Referential Voice (arxiv.org)
	103 points by yu3zhou4 6 days ago \| hide \| past \| pdf \| 110 comments
11.	How good are frontier models at physics? (arxiv.org)
	100 points by qt31415926 17 days ago \| hide \| past \| pdf \| 50 comments
12.	Accurate Models of AMD Matrix Cores (arxiv.org)
	80 points by matt_d 17 days ago \| hide \| past \| pdf \| 11 comments
13.	The Implications of Linguistic Illegibility for LLM Security (arxiv.org)
	79 points by tomjakubowski 15 days ago \| hide \| past \| pdf \| 29 comments
14.	Show HN: Training a model to identify AI web content from structure alone (arxiv.org)
	74 points by jochenmadler 11 days ago \| hide \| past \| pdf \| 28 comments
15.	Procedural Graphs: Self-Evolving Execution Structures for LLM Agents (arxiv.org)
	57 points by omarsar 24 days ago \| hide \| past \| pdf \| 15 comments
16.	GRP-Obliteration: Unaligning LLMs with a Single Unlabeled Prompt (arxiv.org)
	24 points by vital101 18 days ago \| hide \| past \| pdf \| 9 comments
17.	Fixing GRPO's credit assignment problem without evaluating every step (arxiv.org)
	23 points by mrkn1 1 day ago \| hide \| past \| pdf \| 3 comments
18.	Quantized Reasoning Models Think They Need to Think Longer, but They Do Not (arxiv.org)
	13 points by theanonymousone 6 days ago \| hide \| past \| pdf \| 1 comment
19.	Reflections on Trusting Trust, Revisited: Poisoning Self-Modifying AI Coding (arxiv.org)
	13 points by sbulaev 15 days ago \| hide \| past \| pdf \| 1 comment
20.	The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It (arxiv.org)
	11 points by amichail 14 days ago \| hide \| past \| pdf \| 1 comment
21.	Meta$^N$: Recursive Self-Improvement Through Emergent Depth (arxiv.org)
	10 points by Anon84 20 days ago \| hide \| past \| pdf \| 2 comments
22.	Thinking with Looped Flows (arxiv.org)
	9 points by E-Reverance 22 days ago \| hide \| past \| pdf \| 1 comment
23.	RoofLang: Enabling AI-Driven Architecting of LLM Inference Systems (arxiv.org)
	7 points by matt_d 19 days ago \| hide \| past \| pdf \| discuss
24.	Study shows AI is as good as human tutoring for GRE learning gains (arxiv.org)
	6 points by cgn 8 days ago \| hide \| past \| pdf \| 1 comment
25.	KnowBench: Evaluating clinical AI with effort reduction (arxiv.org)
	5 points by kangjl888 18 days ago \| hide \| past \| pdf \| 2 comments
26.	Superhuman AI for Stratego (arxiv.org)
	5 points by droidjj 1 day ago \| hide \| past \| pdf \| 1 comment
27.	A Bitter Lesson for Data Filtering (arxiv.org)
	5 points by nujan_dev 15 days ago \| hide \| past \| pdf \| 1 comment
28.	Inference-Engine Fingerprinting Attacks Are Practical (arxiv.org)
	5 points by sbulaev 13 days ago \| hide \| past \| pdf \| discuss
29.	Looped Transformers as Programmable Computers (2023) (arxiv.org)
	4 points by peter_d_sherman 23 days ago \| hide \| past \| pdf \| 1 comment
30.	Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians (arxiv.org)
	4 points by luispa 3 days ago \| hide \| past \| pdf \| discuss
	More