ML Newsnew | past | best | rss | submit about
Stories from September 18, 2026 (UTC)
Go back a day, month, or year. Go forward a day.
1. An empirical study of harness design for coding agents (arxiv.org)
225 points by wek 16 days ago | hide | past | pdf | 59 comments
2. Cache-to-Cache: Direct Semantic Communication Between LLMs (2025) (arxiv.org)
109 points by rochansinha 16 days ago | hide | past | pdf | 22 comments
3. The Implications of Linguistic Illegibility for LLM Security (arxiv.org)
79 points by tomjakubowski 16 days ago | hide | past | pdf | 29 comments
4. Reflections on Trusting Trust, Revisited: Poisoning Self-Modifying AI Coding (arxiv.org)
13 points by sbulaev 16 days ago | hide | past | pdf | 1 comment
5. The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It (arxiv.org)
8 points by 1317 16 days ago | hide | past | pdf | 1 comment
6. A Bitter Lesson for Data Filtering (arxiv.org)
5 points by nujan_dev 16 days ago | hide | past | pdf | 1 comment
7. Score Centering Stabilizes Off-Policy Reinforcement Learning (arxiv.org)
3 points by zagwdt 15 days ago | hide | past | pdf | discuss
8. SSD-Llama: SSD-Native Inference for Trillion-Parameter Moe on a Consumer PC (arxiv.org)
3 points by AlmostCosmo79 16 days ago | hide | past | pdf | discuss
9. The More It Says, the More You Pay: A Black-Box Audit of Token Inflation in LLM (arxiv.org)
2 points by sbulaev 16 days ago | hide | past | pdf | discuss
10. Alibi: Adversarial Legitimacy Injection in Binaries Against LLM Malware (arxiv.org)
1 point by sbulaev 16 days ago | hide | past | pdf | discuss
11. Much Ado About Noising: Dispelling the Myths of Generative Robotic Control (arxiv.org)
1 point by porridgeraisin 16 days ago | hide | past | pdf | discuss
About | RSS | RSS (all) | HN arXiv