Intel 情报源

OpenAI fought dirty on career-making math problem

OpenAI被指控在数学问题上使用不公平策略。

Muse: Meta's personal AI agent, features and capabilities

Meta的Muse AI代理功能詳述。

Rivian's gambit for full autonomy

Function Arguments Are Not Function Colors

函数参数不施加颜色限制。

Real photos of young girls were in nudify-app ads on Facebook, Instagram

Meta平台上的Nudify应用广告使用未成年人真实照片。

ChatGPT Images 2.5

ChatGPT Images 2.5发布，带来改进。

Generating the P3 Tiling

描述生成P3镶嵌图案的方法。

Flights cancelled at UK airports due to ATC issue

英国机场因航管问题取消航班。

What do Visa and Mastercard do? An intro to card networks

说明Visa和Mastercard的角色与运作…

ReCite: Agentic Reasoning for Faithful Citation

ReCite运用代理推理提升引用准确性。

Procedural Graphs: Self-Evolving Execution Structures for LLM Agents

程序化图谱演化执行结构引导LLM。

Copying explains the collective behavior of AI agents in the wild

AI代理在wiki中的集体行为由…解释。

Entropy-Regularized Rank-Masked Policy Optimization for Test-Time Reinforcement Learning in Code Generation

ERPO通过…将测试时强化学习应用于代码生成。

Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails

同策略专家修正以…提升较弱模型。

ExecCritic: Learn to Test, Test to Improve for Coding Agents

ExecCritic训练代理生成测试并修复代码。

Canonical Color as a Lens into Concept Decodability in Vision Encoders and VLMs

规范色彩可解码性揭示…中的概念信息。

When Does Scale-Invariant Optimization Become Unstable? An Exact Schedule Law with Weight Decay

尺度不变优化稳定性由…控制。

MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents

MeClear运用博弈论清除…中的有害记忆。

SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?

SAEScientist-Bench评估AI代理自主SAE…

Measuring LLM Sycophancy under Sustained Multi-Turn Pressure

SPINE衡量LLM在持续多轮下的谄媚行为。

It's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention

注意力汇集与大量激活源于…

GoDeep: Annotation-Free Open-Vocabulary 3D Scene Understanding via Language-Space Lifting

GoDeep将语言描述提升至3D以…

Everything in Moderation: Per-Domain Coverage Optima and Alignment-Resistant Domain Gaps in Multi-Domain Mid-Training

中期训练领域覆盖最佳值存在，但对齐…

ActReview: Rebuttal-Guided Training Data and Rubric Rewards for Actionable Peer Review Generation

ActReview使用反驳来训练模型以产生可操作的…

ToolLoop: Closed-Loop Tool-Use Data Synthesis via Decomposed Generation and Dynamic Self-Feedback

ToolLoop通过闭环合成工具使用数据…

Performance of Clinical AI System and Physicians and Frontier Language Models in primary care diagnostics

临床AI Doctorina表现优于医师和前沿…

PlayTrain: An Efficient Reinforcement Learning Framework for LLM-Generated Adaptable JavaScript Games

PlayTrain生成JavaScript游戏以进行高效RL…

On the Navier–Stokes Millennium Prize Problem

讨论Navier-Stokes千禧年大奖问题。

Time-Varying Data as Sheaves: an Invitation to Narratives

Narratives框架将时变数据建模为束…

Training-Free Task Vectors for LLM Behavioral Control

引入免训练任务向量以进行LLM行为…

The Audit Decides the Verdict: Instrument Effects Rival Demographic Bias in LLM Decision Audits

审计建构对LLM偏见发现的影响大于…

OUI-1: world's first model for Generative UI

OUI-1是首个生成式UI模型。

Do Reasoning Representations Help Humans Evaluate LLM Outputs?

推理表示不一定有助于人类评估。

Show HN: LLM Attention Visualization

Show HN：可视化LLM注意力的工具。

Answer-Distribution Trajectories: A Stochastic-Dynamics View of LLM Reasoning

追踪LLM答案分布以揭示推理过程。

DOJ Blocked ICE Agent Shooting Charge over Federal Prosecutor's Objections

司法部阻挡ICE探员枪击指控，无视检察官。

It Is Not My Code Anymore

探讨AI辅助代码的所有权与责任。

Let It Go or Learn to Self-Correct: Continuous Diffusion for Constrained Discrete Tasks

连续扩散需自我修正以处理约束。

Connecting the Machines

在网络环境中连接机器。

Evaluation of Contextual Understanding in Large Language Models

提出知识图谱指标S3KG评估LLM。

Deposon: An Auditable, Conservation-Guaranteed, Game-Theoretically Tested Scattering Layer over LLM Reasoning Paths

Deposon为LLM推理添加可审计散射层。

Physics-Informed Deep Learning for False Ventricular Tachycardia Alarm Reduction in the ICU

物理信息深度学习减少ICU假警报。

Y Combinator Early Access Network

Y Combinator推出早期访问网络。

Transformers as In-Context Samplers: From Closed-Form Diffusion to Estimation-Free Sampling

Transformer可作为数据的上下文采样器。

Omni Interaction Agent Technical Report

Gander统一全感知、实时互动等。

Good Pretraining, Bad SFT: Checkpoint Quality Across the Training Stack

最佳预训练检查点在完整SFT后未必最佳。

PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous Driving

PlannerForge使用LLM代理进行基于场景的ADS测试。

Do you think it happened? Research stolen from their Codex private chats

指控从私人聊天中窃取研究。

SQLMorph: Query Mutation and Fine-Grained Metrics for Text-to-SQL Evaluation

SQLMorph变异查询并提供细粒度指标…

LG TVs caught spying even when offline or on standby

LG电视据报即使关机或闲置也会监视。

ONE CYLinder: A Benchmark for Graph-Based Surrogate Modeling of Unsteady Bluff-Body Flows

用于图形代理建模的ONECYL基准…

SkillAdam: Stable and Efficient Skill Evolution for Agents

SkillAdam以Adam启发的方式优化代理技能…

Evaluating and Improving Evidence-Grounded Fact-Checking in LLMs via Multi-Round Evidence Ablation

LLM在…中依赖参数知识而非证据。

AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing

AuK是语音生成和…的开源模型。

When Models Defer to Wrong Answers: A Robustness Audit of Source-Attributed Cues in Multiple-Choice QA

来源归因提示可使LLM服从错误答案。

C*: Unifying Programming and Verification in C

C*语言结合编程与形式验证。

Into the depths of C: Elaborating the de facto standards (2016)

探讨2016年起的实际C标准。

Experience Funnel: A State-Policy Alternating Loop for Self-Evolving Agents

自演化代理交替快速状态和慢速策略…

US police fear Meta smart glasses could be used to secretly record them

警方担忧Meta眼镜可秘密录像。

The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning Evaluation

BatchNorm伪影混淆遗忘评估…