September 29, 2026 front

ML Newsnew | past | best | rss | submit

Stories from September 29, 2026 (UTC)

Go back a day, month, or year. Go forward a day.

1.	Self-Play Pretraining with Zero Data (arxiv.org)
	3 points by 7777777phil 4 days ago \| hide \| past \| pdf \| discuss
2.	Schedules Are Solvable Symbols: Tuning-Free Compilation of Tile Programs (arxiv.org)
	2 points by matt_d 4 days ago \| hide \| past \| pdf \| discuss
3.	Netflix replaced their recommendation algorithm with an LLM (arxiv.org)
	2 points by nreece 4 days ago \| hide \| past \| pdf \| 1 comment
4.	Shockingly Simple Self-retrospection Improves Agentic Models Without RL (arxiv.org)
	2 points by Betelbuddy 4 days ago \| hide \| past \| pdf \| discuss
5.	Language Models Act on Hidden Valence (arxiv.org)
	2 points by sbulaev 4 days ago \| hide \| past \| pdf \| discuss
6.	Pluralis Towards a Multicultural Multimodal, Multilingual Benchmark for AI Risk (arxiv.org)
	1 point by thinkevolve 4 days ago \| hide \| past \| pdf \| discuss
7.	AmpleGCG: Learning a Universal Generative Model for Jailbreaking (arxiv.org)
	1 point by Anon84 4 days ago \| hide \| past \| pdf \| discuss
8.	Learning How to Forget: Fine-Tuning for Long-Context Sparse Attention (arxiv.org)
	1 point by theanonymousone 4 days ago \| hide \| past \| pdf \| discuss
9.	Fathom: Per-query read depth for sparse decoding over offloaded KV caches (arxiv.org)
	1 point by vivekkalyanaran 4 days ago \| hide \| past \| pdf \| discuss
10.	User Model Extraction via Belief Self-Distillation (arxiv.org)
	1 point by sbulaev 4 days ago \| hide \| past \| pdf \| discuss