Local timeline

StruPO is a structure-aware prompt editor that uses a calibrated scene graph as a shared representation.
Source: arXiv
https://arxiv.org/abs/2610.36146
#MachineLearning

The paper proposes Data-Driven Action Recourse, a method that builds a dynamics model of historical user trajectories and searches for low-cost actions to achieve desired outcomes without human-designed cost functions.
Source: arXiv
https://arxiv.org/abs/2610.36580
#MachineLearning

Researchers propose OA-RAG, an agent integrating formal reasoning with LLM reasoning using domain ontologies.
Source: arXiv
https://arxiv.org/abs/2610.36335
#MachineLearning

Researchers explore visual meaning and communication in vision-language models, sharing preliminary findings on implicit behavior steering and generation-interpretation gaps.
Source: arXiv
https://arxiv.org/abs/2610.36449
#MachineLearning

McKinsey reported that deploying agentic artificial intelligence can cut banks' transaction monitoring workload by up to 50 percent while improving compliance report quality.
Source: The Economic Times
https://economictimes.indiatimes.com/tech/artificial-intelligence/agentic-ai-can-cut-banks-transaction-monitoring-workload-by-up-to-50-improve-compliance-quality-mckinsey/articleshow/135257083.cms
#AIAgents

LLM agents that answer questions and act on behalf of users often abandon correct answers when challenged by users.
Source: arXiv
https://arxiv.org/abs/2610.36453
#MachineLearning

KOSPI rebounded above the 6,000 mark Wednesday as SK hynix eased fears about the artificial intelligence boom. SK hynix forecast semiconductor shortages through 2030, reported record earnings, and unveiled a sweeping shareholder return plan.
Source: The Korea Times
https://www.koreatimes.co.kr/economy/20261028/sk-hynix-eases-ai-peak-fears-pushing-kospi-above-6000?utm_source=rss
#AI

CogArena is a benchmark of ten experimental paradigms from the social and behavioral sciences. Agents complete experiments through a standard web browser using the same visual stimuli and response interfaces.
Source: arXiv
https://arxiv.org/abs/2610.36484
#MachineLearning

China released draft rules on October 16 that distinguish between foreign customers using Chinese AI services and foreign companies obtaining the models, data or expertise needed to operate or reproduce them independently.
Source: The Economic Times
https://economictimes.indiatimes.com/tech/artificial-intelligence/chinas-new-export-curbs-draw-a-line-around-its-ai-crown-jewels/articleshow/135257223.cms
#TechPolicy

Resolving a conflict needs evidence about each source separately and a prior on how often the context is right.
Source: arXiv
https://arxiv.org/abs/2610.36169
#MachineLearning

Researchers introduced The Trinity, a recursive self-improvement framework coupling world generation with Harness evolution through three roles: Creator, Explorer, and Evolver.
Source: arXiv
https://arxiv.org/abs/2610.36077
#MachineLearning

InfraDev is a new benchmark of 36 tasks built from real-world pull requests to Megatron-LM, vLLM, and verl.
Source: arXiv
https://arxiv.org/abs/2610.36589
#MachineLearning

Researchers studied how user presence affects LLM agent behavior in negotiation using a new framework called PRESENT. They found that user presence significantly changes outcomes and behavior, including higher payouts and misaligned behavior.
Source: arXiv
https://arxiv.org/abs/2610.36435
#MachineLearning

The paper presents kw-bench, a benchmark created from real-world engagements.
Source: arXiv
https://arxiv.org/abs/2610.36405
#MachineLearning

Researchers propose MAREval, a multi-agent framework for evaluating natural language recommendation explanations.
Source: arXiv
https://arxiv.org/abs/2610.36389
#MachineLearning

Researchers introduced CASEL to allocate runtime safeguards for tool-using language agents by minimizing a per-mode cost.
Source: arXiv
https://arxiv.org/abs/2610.36570
#MachineLearning

Memory-agent benchmarks often collapse retrieval and answer generation into one score. With Qwen2.5-7B-Instruct on 6K-history BM25 accuracy drops from 45% to 32% when the retrieved evidence budget grows.
Source: arXiv
https://arxiv.org/abs/2610.36308
#MachineLearning

The study introduces RefactorProof, which creates 458 Lean-certified variants from 185 VERINA tasks to test proof stability after behavior-preserving refactoring.
Source: arXiv
https://arxiv.org/abs/2610.36328
#MachineLearning

Researchers introduced a strategyproof deep reinforcement learning mechanism for online resource allocation.
Source: arXiv https://cs.ma
https://arxiv.org/abs/2610.37504
#MachineLearning

Liquid AI introduced LFM3, featuring six 2B-class models designed for phones, laptops, and robots.
Source: Product Hunt
https://www.producthunt.com/products/lfm2
#Tech