今日必看

Carryover Drafting: Recycling Rejected States for Speculative Decoding

arXiv:2609.14717v1 Announce Type: cross Abstract: Speculative decoding accelerates LLM inference by verifying multiple drafted tokens in parallel, allowing a single target forward pass to accept several tokens. By construction, verification computes representations for both accep

ECAS: An Edge-Controlled Agentic System for Validation-Gated Scientific Application Execution

arXiv:2609.14211v1 Announce Type: cross Abstract: Scientific applications increasingly rely on high-performance computing (HPC), yet translating a scientist's high-level goal into a correct target-scale execution remains brittle and labor-intensive. Large language model (LLM) age

EMR: Self-Evolving Medical Multi-Agent System via Experience Mining and Reuse

arXiv:2609.15161v1 Announce Type: cross Abstract: Large language model (LLM) driven multi-agent systems have shown promise in complex clinical reasoning, yet existing approaches rely on static strategies and lack persistent clinical memory, preventing self-evolving from prior dia

Grounded Adjudication of Variations across Extracted TimeLines (GAVEL): Comparing Clinical Timelines Against Their Case Reports

arXiv:2609.13475v1 Announce Type: new Abstract: Existing pipelines for clinical timeline extraction from case reports are evaluated using an expert reference and are limited by imperfect reference annotations and imprecise event alignment. We developed GAVEL, an LLM judge protoco

K-Bench: A Benchmark for LLM Unlearning in Agentic Deployments

arXiv:2609.12808v2 Announce Type: replace Abstract: Unlearning benchmarks such as TOFU and MUSE certify forgetting by reading the model's final answer, where a model that refuses to answer already counts as having forgotten. We show that this model-level certificate does not tran

K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations

arXiv:2609.15855v2 Announce Type: cross Abstract: People increasingly use large language models (LLMs) for mental health support, yet their safety in evolving, high-risk conversations remains poorly characterised. We developed K-Bench, a clinician-calibrated, protected benchmark

KREL: Automatic Medical Coding via Knowledge-Guided Reasoning over Clinical Evidence with LLMs

arXiv:2608.20887v2 Announce Type: replace-cross Abstract: Automatic Medical Coding (AMC), which assigns standardized International Classification of Diseases (ICD) codes to clinical notes, is essential for medical reimbursement, quality reporting, and clinical research. Existing

OrchSLM: Probing the Dynamics of Small Language Model Orchestration

arXiv:2609.13470v1 Announce Type: new Abstract: Although large language models (LLMs) have demonstrated remarkable capabilities, their reliance on cloud-scale infrastructure poses fundamental challenges for deployment in agentic pipelines, including latency, privacy, connectivity

Positioning manuscripts in the scientific landscape with agentic AI

arXiv:2609.13760v1 Announce Type: new Abstract: Publishing a research manuscript is a routine yet demanding part of scientific life: time-consuming, stressful, and often uncertain in outcome. Recent advances in large language model (LLM)-based agentic AI have shown promise across

REGEN: Replay-recycling for Expert-to-Generalist distillation with Offline Reinforcement Learning

arXiv:2607.19450v3 Announce Type: replace-cross Abstract: Large-scale online reinforcement learning (RL) is the predominant means of eliciting advanced abilities including long-term reasoning and agentic tool use in large language models (LLMs). However, continuing to scale it ac

Route, Don't Fix: Regime-Dependent Decoding Correction and a Trajectory-Gated Router for Reliable Clinical LLM Answer Selection

arXiv:2609.14825v1 Announce Type: cross Abstract: Large language models (LLMs) are often deemed unsafe for clinical question answering because of their tendency to hallucinate. Retrieval augmentation, fine-tuning, and external verifiers require new infrastructure that clinical go

SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric Scheduling

arXiv:2607.08565v2 Announce Type: replace-cross Abstract: LLM scheduling is critical to serving, yet how well existing designs fit agentic serving--where agents, not humans, issue the requests--remains unclear. Agents shift the workload in two ways: they consume many more tokens

Same Patient, Different Order: Action-Level Reliability of Clinical LLM Agents Under Repeated Runs

arXiv:2609.13582v1 Announce Type: cross Abstract: A clinical agent benchmark can report the same verdict on identical inputs while the agent files a materially different order on each run. Such agents order tests, request medications and place referrals, yet benchmarks typically

SemVerBench: Benchmarking LLM Comprehension of Version-Constraint Resolution Semantics

arXiv:2609.11180v2 Announce Type: replace Abstract: Large language model (LLM) coding agents constantly decide whether a version satisfies a constraint such as ^1.2.3 or >=2.0, 1.2 means >=1.3.0) traps every model on Cargo (near 60%), and although standard PEP 440 prefix matching

Vibe Patenting: Evaluating LLM Judges for Professional Patent-Drafting Agents

arXiv:2609.13422v1 Announce Type: new Abstract: LLM judges are increasingly used to evaluate and improve AI-generated outputs, yet their reliability for complex professional work remains unclear. We study this problem through Vibe Patenting, an end-to-end patent-drafting testbed

VideoScout: Learning Agentic Active Exploration with Adaptive Reasoning Pacing for Long Video Understanding

arXiv:2609.15606v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) have achieved remarkable progress on short video understanding yet remain limited on long videos due to the limited visual context window. Prevailing approaches rely on uniform frame sampli

When Malicious Instructions Persist: Persistent Memory Poisoning Attack on Harness-Based Agents

arXiv:2609.13889v1 Announce Type: cross Abstract: Harness design has transformed the development of LLM-based agents by integrating memory, tool use, and runtime control. However, this design also introduces security and privacy risks because malicious instructions from external

How workers are unlocking new ways of working

New OpenAI Economic Research shows how workers use AI beyond traditional roles and which new activities become recurring parts of their work.

Code scanning AI Scan no longer requires CodeQL default setup

You can now use AI Scan for pull requests to find security vulnerabilities, even when CodeQL default setup isn’t enabled on a repository. Previously, AI Scan for pull requests only ran on repositories where CodeQL default setup was configured. What’s changed Code scanning and AI

Anthropic 将 Claude Cowork 与聊天合并为统一的 Claude，并推出 Docs、Slides 等功能

Anthropic 宣布 Claude Cowork 与聊天合并为一个 Claude，任务无需再选择入口，Cowork 和 Design 的能力可在任意对话中使用，未来几周内向 Pro 和 Max 计划推出。 🔗 阅读原文 via AIHOT · https://aihot.news/items/cmu4c8qar0hzero4w3b6n1jyf

SpaceXAI 直播 72 小时用 Grok Bot 从零造公司

📌 一句话摘要 作者介绍 SpaceXAI 将于 9 月 15—17 日在旧金山直播，由三人用 Grok Bot 在 72 小时内完成从构思、商业计划到工程部署的全链路，全程未剪辑公开。 📝 详细摘要 作者介绍 SpaceXAI 的一场公开压力测试：9 月 15 日到 17 日，Matt Palmer、Lauren Tan 和 Roshan Sadanani 在旧金山 The Howard 现场直播，用 Grok Bot 跑完从构思、商业计划、产品决策到工程开发和部署的全链路，每天早 8:30 到晚 6:00 PT 免费观看。日程为第一天 Grok Bo

How good are frontier models at physics?

Accurate Models of AMD Matrix Cores

Your AI agents can now control your Google Home devices

Google is launching early access to a new MCP server for Google Home, allowing AI agents like Claude, ChatGPT, and others to control connected devices, review camera summaries, and access smart home activity using natural language.