August 19, 2026 front

ML Newsnew | past | best | rss | submit

Stories from August 19, 2026 (UTC)

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

1.	Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces (2025) (arxiv.org)
	316 points by nunodonato 46 days ago \| hide \| past \| pdf \| 281 comments
2.	Chain-of-Thought Reasoning in the Wild Is Not Always Faithful (2025) (arxiv.org)
	66 points by florianherrengt 46 days ago \| hide \| past \| pdf \| 38 comments
3.	Improving the Matrix Multiplication Exponent (arxiv.org)
	6 points by aaraujo002 46 days ago \| hide \| past \| pdf \| discuss
4.	Adding Error Bars to Evals: A Statistical Approach to Language Model Evaluations (arxiv.org)
	3 points by Anon84 46 days ago \| hide \| past \| pdf \| discuss
5.	Benchmarking General-Purpose Agents on Market-Validated End-to-End Workflows (arxiv.org)
	3 points by tcp_handshaker 46 days ago \| hide \| past \| pdf \| discuss
6.	Fragility of Self-Improving Agents: Variance, Task Order, and Underspecification (arxiv.org)
	3 points by tcp_handshaker 46 days ago \| hide \| past \| pdf \| discuss
7.	Model Hypnosis: Strong control of AI via additive subliminal effects (arxiv.org)
	3 points by sbulaev 46 days ago \| hide \| past \| pdf \| discuss
8.	Stealing Reasoning Traces from Proprietary LLM APIs (arxiv.org)
	2 points by amai 46 days ago \| hide \| past \| pdf \| 1 comment
9.	Sampling More, Getting Less: Calibration Is the Diversity Bottleneck in LLMs (arxiv.org)
	2 points by clukic 46 days ago \| hide \| past \| pdf \| discuss
10.	AutoResearch: Insight In, Hallucination Out (arxiv.org)
	2 points by johnbarron 46 days ago \| hide \| past \| pdf \| discuss