new

Get trending papers in your email inbox once a day!

Get trending papers in your email inbox!

Daily Papers

byAK and the research community

Oct 1

Submitted by

fangwu97

UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement

stanfordnlp

Submitted by

LH2101

The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation

·
7 authors

Submitted by

MarkShaw99

Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence

amazon

Submitted by

chenmeijia30

False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents

RutgersU

Rutgers University

Submitted by

lz1001

AREX-2: Advancing Self-Improving Agents through Long-Horizon Reflective Tasks

BAAI

Beijing Academy of Artificial Intelligence

Submitted by

Nardien

Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents

nvidia

5

Submitted by

passing2961

EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving for Scientific Discovery

minnesotanlp

Submitted by

ziyjiang

WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents

UCSantaBarbara

University of California, Santa Barbara

Submitted by

chengq9

Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI

apodex

Submitted by

WenyiWU0111

RSIGame: Autonomous Agentic Game Development with Recursive Self-improvement

RSIGame

Submitted by

Gnonymous

EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

SJTU

Shanghai Jiao Tong University

Submitted by

ElvisWang111

Thinking Outside the Box: Can Language Models Rely on External Guidance Selectively?

SouthernUniversityofScienceandTechnology

Southern University of Science and Technology

Submitted by

HwanChang0106

Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering

kaist-ai

Submitted by

zhiyuan218

More Choices, Fewer Decisions: Ordinal-Scale Bias in JEV-like Direct-Decision Models

JilinUniversity001

Jilin University

Submitted by

imhgchoi

AIM: Agentic Idea Management for Automated Research

google

3

Submitted by

Leozkl

Agent Error Dataset: Scaling 50,000 Error--Diagnosis Pairs for Failure Analysis and Error-Aware Post-Training

apodex

Submitted by

youuor7r

Unmask the State: When Does State Adaptation Matter for Masked Diffusion Language Models

HOLILAB

Submitted by

pashocles

LANTERN: Illuminating Hidden Mathematical Knowledge in Language Models

·
5 authors

2

Submitted by

naghamo

It's Not What the Image Shows: Irrelevant Context Destabilises VLM Judges Without Informing Them

Technion

Technion Israel institute of technology

Submitted by

henry-y1

DC-SAE: Deep Compression Semantic Autoencoder for Faster Diffusion Convergence

DAGroup-PKU

Submitted by

chengkunli

TERRA: Terrain-Aware Reconstruction, Retargeting and Control for Musculoskeletal Locomotion

amathislab

Mathis Group @ EPFL

Submitted by

meravkeidar

DyRAD: Radar Novel View Synthesis for Dynamic Driving Scenes

Technion

Technion Israel institute of technology

Submitted by

AllenJin

Breaking Babel: A Self-Evolving Multi-Agent System for Long-Form Subtitle Translation

amazon

2

Submitted by

iLOVE2D

OSWorld-Science: A Benchmark of Computer Use Agents for Learning and Using Scientific Software

SciAILab

Submitted by

qizekun

Systematically Exploring the Capabilities of GPT-6 Astra as Embodied Policies

Galbot

Submitted by

Gtime666

LoopVL: Recurrent Visual Intelligence

RUC

Renmin University of China

Submitted by

donghao-zhou

ThinkV2V: Unleashing the Reasoning Capability of MLLMs for Instruction-Guided Video Editing

·
13 authors

Submitted by

Olivia121

BiasReducer: Adaptive Bias Mitigation for Reward Models

CarnegieMellonU

Carnegie Mellon University

Submitted by

ahatamiz

PivotOPD: Learning to Recover from Pivotal Mistakes in Multi-Turn Agents

nvidia

Submitted by

yanbeic

Scaling Laws for Looped Mixture of Experts

2

Submitted by

Jongwondd

The Low-Rank Structure of VLA Reinforcement Learning

HOLILAB

2

Submitted by

nikosusc

MILO: Automated Harness Discovery via Orchestrated Multi-Agent Evolution

amazon

2

Submitted by

kingofspace0wzz

CUA-SWE: When Computer-Use Agents Meet Visual Software Engineering

CarnegieMellonU

Carnegie Mellon University

Submitted by

LukasKnobel

I Have a Stream: Making Self-Supervised Learning Work on Continuous Video

FunAILab

Fundamental AI Lab at UTN

2

Submitted by

JEdward

RoboCoach: World Models as Active Coaches for Compositional Robot Skills

Tsinghua University

Submitted by

Glanceyes

A2Z GameSpec-Bench: How Faithfully Can Coding Agents Generate Games from Game Design Specifications?

KRAFTON

Submitted by

henggg

DuoOPD: Learning from Joint Teacher-Student Outcomes for Multi-Task On-Policy Distillation

·
9 authors

Submitted by

Miladyz

Rubric Rewards from Item Response Theory

microsoft

Submitted by

atsuki-yamaguchi

Synthetic Pre-pretraining Survives Scale, but Not as a Grammatical Prior

Submitted by

taesiri

Physis-Lang: Self-Evolving Language as a Physical Representation for Video World Model

nvidia

Submitted by

puar-playground

Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free Voice Cloning

·
3 authors

1

Submitted by

CongWei1230

PixelUMM: Encoder-Free Unified Image and Video Understanding and Generation

nvidia

Submitted by

guanqun-yang

PatchHolmes: Agentic Patch Retrieval via Listwise Selection

StevensAGI

Stevens Institute of Technology

Submitted by

rish-1086

AdviSD: Learning to Advise Frontier LLMs via Targeted Multi-Turn Self-Distillation

google

1

Submitted by

Huxaifa

Safety of Latent Communication in Multi-Agent Systems

cispa

CISPA – Helmholtz-Zentrum für Informationssicherheit gGmbH

Submitted by

tzt

The Evolution of Attention in Large Language Models: Mechanisms, Trade-offs, and Emerging Trends

·
6 authors

2

Submitted by

ilgee

Training LLM Judges from Language Feedback via Position-Selective Self-Distillation

amazon

Submitted by

nikopavl

SeLMRoute: Probabilistic Semantic Evidence for Large Language Model Routing

Indigma

Indigma Innovations

Submitted by

billzhao1030

NavHarness: Towards Lifelong Embodied Navigation

·
9 authors

Submitted by

ismail31415

Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning

kaist-ai

Submitted by

Luli3220

Not Every Token Is Worth Distilling: Selective Supervision for Direct-OPD

ECNU

East China Normal University

Submitted by

softmax

WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms

ISTA-DASLab

IST Austria Distributed Algorithms and Systems Lab

2

Submitted by

Yesianrohn

Decompose Radicals, Then Reward: Fine-Grained Inspection for Accurate Chinese Text Rendering

Fudan-University

Fudan University

1

Submitted by

latentdulhan

Removing Timing Shortcuts Improves Non-Invasive Brain-to-Text

pnpl

Parker Jones Neural Processing Lab (PNPL)

Submitted by

reasonwang

SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation

·
5 authors

Submitted by

fengnian1678

Beyond the Remembered World: Predictive 4D Belief for Persistent Navigation in Evolving Worlds

Submitted by

longphann

CheatBench: Measuring Reward Gaming in AI Agents

cais

Center for AI Safety

Submitted by

guanqun-yang

SkillSeek: Revisiting Agent Skill Retrieval at Marketplace Scale

StevensAGI

Stevens Institute of Technology

Submitted by

HeiXiong620

CoEvoWhen: Policy-Tool Coevolution for Ultra-Long Video Temporal Grounding

zju

Zhejiang University

Submitted by

taesiri

Game-Guided Skill Discovery through Self-Play for Playable Agent Control

·
4 authors

Submitted by

lorenzomazza

Understanding Multimodality in Generative Behavioral Cloning

nct-tso

Department of Translational Surgical Oncology (TSO) at NCT Dresden

Submitted by

rzdiversity

See it, Say it, Sorted: Mechanistic Diagnosis and Parameter-Space Mitigation of Emergent Misalignment in LLMs

UniversityofBirmingham

University of Birmingham

Submitted by

NDPP

BIABench: Evaluating AI agents on real-world bioimage analysis tasks

notredame

University of Notre Dame

Submitted by

hanwenliu6825

DAGent: Evaluate-then-Grow Planning for Deep Research Agents

newyorkuniversity

New York University

Submitted by

rafiibnsultan

SpatialCORE: Confidence-Aware Grounded Spatial Reasoning in Large Vision--Language Models

·
7 authors

Submitted by

jisx

Label-free steering: Compressing test-time reinforcement learning into bias-only subspaces

OLAResearchX

Omni Language AI Research

Submitted by

dipankarsarkar

Evaluating Bounded Autonomy in Regulated Agentic AI: A Diagnostic Harness with Constitutional Rewards, Escalation Labels, and Runtime Governance

·
1 authors

1

Submitted by

mario-sanz

Dating the Model: Hidden Dates in System Prompts Affect LLM Evaluation

·
4 authors

1

Submitted by

bunny127

EviRover: Reinforcing Agentic Perception Beyond a Glance

·
7 authors

Submitted by

awang3382

Aligning One-Step Generative Models with Reward-Weighted Transport Distillation

ByteDance-Seed

Submitted by

alesspalma

Prompted Identity Degrades Cooperation in Multi-Agent LLM Systems

·
5 authors

1

Submitted by

shancn7

Who Said What, and Will It Be Remembered? Evaluating Persistent Speaker Attribution Across Meetings

·
3 authors

2

Submitted by

gzxiong

MemLife: Curating and Reasoning over Long-Term Egocentric Video Memories

meta

2

Submitted by

Wanux

Mitigating the Length-Scaling Tax with Online Distillation

·
4 authors

Submitted by

jjrussell10

How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text

pangram

Submitted by

juyil

LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception

·
12 authors

Submitted by

Muhansp

The Endless Exam: Mathematical Constructions from Today's Models toward Superintelligence

·
1 authors

Submitted by

greycat111

ATLAS: Aligned Transport of Latent Structure for Reliable World Model Planning

Pennsylvania State University

Submitted by

mbur

Retrieval Capacity of Self-Attention Under Competition

·
3 authors

Submitted by

mbur

The Geometry of Inference in Transformer Residual Streams

·
3 authors

Submitted by

Regiayoung

SlideDP: Scaling Host-Resident LLM Fine-Tuning Across Multiple GPUs

·
8 authors

Submitted by

Corbenic

Working Around the Compute Ceiling: Byte-Exact Memory in Galahad Makes LLM Reading a One-Time Cost LLM Reading a One-Time Cost

Corbenci

Submitted by

brucelyu

Decision-Oriented Recommendation Reranking: An Empirical Study of Jev

SingaporeManagementUniversity

Singapore Management University

Submitted by

bhavikmangla

Almost Human, Except When It Matters: VoxParity and the Decisions a Voice Should Change

·
1 authors

Submitted by

liuyilun2000

Learning Chaos Without Seeing Chaos: Extrapolation of Global Dynamics in Autoregressive Transformers

LMU

Ludwig Maximilian University of Munich

Submitted by

rafiibnsultan

Soft Spatial Reasoning

·
7 authors