Search: “models”

Reasoning Compression with Mixed-Policy Distillation

arXiv:2605.08776v1 Announce Type: new Abstract: Reasoning-centric large language models (LLMs) achieve strong performance by generating intermediate reasoning trajectories, but often incur excessive t

safetyarxiv-cs-ai

Robust Server Defense Against Unreliable Clients in One-Shot Fair Collaborative Machine Learning

DGX agent

arXiv:2605.08616v1 Announce Type: new Abstract: Collaborative machine learning (CML) enables multiple clients to train a global model jointly in a data-distributed setting. To address data privacy and

model-releasesarxiv-cs-lg

12 May 2026

ROM: Real-time Overthinking Mitigation via Streaming Detection and Intervention

arXiv:2603.22016v2 Announce Type: replace-cross Abstract: Large Reasoning Models (LRMs) often reach a correct solution before their long Chain-of-Thought trace ends, yet continue with redundant verifi

Route Before Retrieve: Activating Latent Routing Abilities of LLMs for RAG vs. Long-Context Selection

model-releasesarxiv-cs-cl

arXiv:2605.10235v1 Announce Type: new Abstract: Recent advances in large language models (LLMs) have expanded the context window to beyond 128K tokens, enabling long-document understanding and multi-s

S2FT: Parameter-Efficient Fine-Tuning in Sparse Spectrum Domain

arXiv:2605.08589v1 Announce Type: new Abstract: Parameter Efficient Fine-Tuning (PEFT) is a key technique for adapting a large pretrained model to downstream tasks by fine-tuning only a small number o

SayNext-Bench: Why Do LLMs Struggle with Next-Utterance Anticipation?

arXiv:2602.00327v2 Announce Type: replace Abstract: We explore the use of large language models (LLMs) for next-utterance anticipation in human dialogue. Despite recent advances in LLMs demonstrating

SciVQR: A Multidisciplinary Multimodal Benchmark for Advanced Scientific Reasoning Evaluation

arXiv:2605.10187v1 Announce Type: new Abstract: Scientific reasoning is a key aspect of human intelligence, requiring the integration of multimodal inputs, domain expertise, and multi-step inference a

SeePhys Pro: Diagnosing Modality Transfer and Blind-Training Effects in Multimodal RLVR for Physics Reasoning

arXiv:2605.09266v1 Announce Type: new Abstract: We introduce SeePhys Pro, a fine-grained modality transfer benchmark that studies whether models preserve the same reasoning capability when critical in

SleepWalk: A Three-Tier Benchmark for Stress-Testing Instruction-Guided Vision-Language Navigation

arXiv:2605.10376v1 Announce Type: new Abstract: Vision-Language Models (VLMs) have advanced rapidly in multimodal perception and language understanding, yet it remains unclear whether they can reliabl

SlimSpec: Low-Rank Draft LM-Head for Accelerated Speculative Decoding

arXiv:2605.10453v1 Announce Type: cross Abstract: Speculative decoding speeds up autoregressive generation in Large Language Models (LLMs) through a two-step procedure, where a lightweight draft model

researcharxiv-cs-cl

Steerable but Not Decodable: Function Vectors Operate Beyond the Logit Lens

arXiv:2604.02608v2 Announce Type: replace Abstract: Activation steering presupposes that task-relevant behaviors correspond to linear directions in activation space -- directions that should both stee

SWE Atlas: Benchmarking Coding Agents Beyond Issue Resolution

arXiv:2605.08366v1 Announce Type: new Abstract: We introduce SWE Atlas, a benchmark suite for coding agents spanning three professional software engineering workflows: Codebase Q&A (124 tasks), Test W

TAD: Temporal-Aware Trajectory Self-Distillation for Fast and Accurate Diffusion LLM

arXiv:2605.09536v1 Announce Type: cross Abstract: Diffusion large language models (dLLMs) offer a promising paradigm for parallel text generation, but in practice they face an accuracy-parallelism tra

researcharxiv-cs-ai

Agents

Temporal Sampling Frequency Matters: A Capacity-Aware Study of End-to-End Driving Trajectory Prediction

arXiv:2605.10388v1 Announce Type: new Abstract: End to end (E2E) autonomous driving trajectory prediction is often trained with camera frames sampled at the highest available temporal frequency, assum

agentsarxiv-cs-cv

The Differences Between Direct Alignment Algorithms are a Blur

arXiv:2502.01237v3 Announce Type: replace Abstract: Direct Alignment Algorithms (DAAs) simplify LLM alignment by directly optimizing policies, bypassing reward modeling and RL. While DAAs differ in th

To Redact, or not to Redact? A Local LLM Approach to Deliberative Process Privilege Classification

arXiv:2605.10211v1 Announce Type: cross Abstract: Government transparency laws, like the Freedom of Information (FOIA) acts in the United States and United Kingdom, and the Woo (Open Government Act) i

Tokens-per-Parameter Coverage Is Critical for Robust LLM Scaling Law Extrapolation

arXiv:2605.08541v1 Announce Type: new Abstract: Neural scaling laws approximate a language model's loss as a power-law function of parameter count N and token count D. Following Chinchilla-style compu

Towards Cross-lingual Values Judgment: A Consensus-Pluralism Perspective

arXiv:2602.17283v2 Announce Type: replace-cross Abstract: As large language models (LLMs) are employed worldwide, existing evaluation paradigms for their multilingual capabilities primarily focus on f

Two Ways to De-Bias an LLM-as-a-Judge: A Continuous-Score Comparison of Hierarchical Bayesian Calibration and Neural-ODE Score Transport

model-releasesarxiv-cs-cl

arXiv:2605.09227v1 Announce Type: new Abstract: [Abridged] Using a Large Language Model (LLM) as an automatic rater (LLM-as-a-judge) is cheap but potentially biased: some judges run lenient, others st

UFO: A Unified Flow-Oriented Framework for Robust Continual Graph Learning

arXiv:2605.09862v1 Announce Type: cross Abstract: Graph learning research has increasingly shifted toward continual graph learning (CGL), which better reflects real-world scenarios where graphs evolve

UMEDA: Unified Multi-modal Efficient Data Fusion for Privacy-Preserving Graph Federated Learning via Spectral-Gated Attention and Diffusion-Based Operator Alignment

arXiv:2605.08288v1 Announce Type: cross Abstract: Device-free localization trains models from heterogeneous wireless and visual sensors (e.g., Wi-Fi, LiDAR) distributed across edge devices. Federated

What should post-training optimize? A test-time scaling law perspective

arXiv:2605.10716v1 Announce Type: new Abstract: Large language models are increasingly deployed with test-time strategies: sample N responses, score them with a reward model or verifier, and return th

safetyarxiv-cs-lg

When Attention Beats Fourier: Multi-Scale Transformers for PDE Solving on Irregular Domains

arXiv:2605.08318v1 Announce Type: cross Abstract: We study the problem of architecture selection for deep learning models trained to solve partial differential equations (PDEs), asking when transforme

When is the last time a general purpose LLM (putting aside hybrid systems like Claude Code with special purpose symbolic harnesses) last com…

model-releasesgary-marcus--x

When is the last time a general purpose LLM (putting aside hybrid systems like Claude Code with special purpose symbolic harnesses) last completely blew away all competing prior models? GPT-4 relative

WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors

arXiv:2605.10434v1 Announce Type: new Abstract: Commercial video generation systems such as Seedance2.0 and Veo3.1 have rapidly improved, strengthening the view that video generators may be evolving i

A Theory of Online Learning with Autoregressive Chain-of-Thought Reasoning

arXiv:2605.06819v1 Announce Type: new Abstract: Autoregressive generation lies at the heart of the mechanism of large language models. It can be viewed as the repeated application of a next-token gene

researcharxiv-cs-lg

Activation Differences Reveal Backdoors: A Comparison of SAE Architectures

arXiv:2605.07324v1 Announce Type: cross Abstract: Backdoor attacks on language models pose a significant threat to AI safety, where models behave normally on most inputs but exhibit harmful behavior w

safetyarxiv-cs-ai

Agentick: A Unified Benchmark for General Sequential Decision-Making Agents

arXiv:2605.06869v1 Announce Type: new Abstract: AI agent research spans a wide spectrum: from RL agents that learn from scratch to foundation model agents that leverage pre-trained knowledge, yet no u

Architecting a resilient, scalable and secure foundation for the agentic era

model-releasesgoogle-cloud-ai

Across the public sector, the conversation has shifted; we are no longer just talking about the potential of AI, we are already seeing the impact. While visionary leadership and cultural buy-in are cr

BEAVER: An Efficient Deterministic LLM Verifier

arXiv:2512.05439v2 Announce Type: replace Abstract: As large language models (LLMs) transition from research prototypes to production systems, practitioners often need reliable methods to verify model

safetyarxiv-cs-ai

Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation

arXiv:2605.07111v1 Announce Type: cross Abstract: Recent literature on fine-tuning Large Language Models highlights a fundamental debate. While Full Fine-Tuning (FFT) provides the representational pla

Beyond the Black Box: Interpretability of Agentic AI Tool Use

arXiv:2605.06890v1 Announce Type: new Abstract: AI agents are promising for high-stakes enterprise workflows, but dependable deployment remains limited because tool-use failures are difficult to diagn

CarCrashNet: A Large-Scale Dataset and Hierarchical Neural Solver for Data-Driven Structural Crash Simulation

arXiv:2605.07098v1 Announce Type: new Abstract: Crash simulation is a cornerstone of modern vehicle development because it reduces the need for costly physical prototypes, accelerates safety-driven de

CONSIGN: Conformal Segmentation Informed by Spatial Groupings via Decomposition

arXiv:2505.14113v3 Announce Type: replace Abstract: Most machine learning-based image segmentation models produce pixel-wise confidence scores that represent the model's predicted probability for each

researcharxiv-cs-cv

CrossCult-KIBench: A Benchmark for Cross-Cultural Knowledge Insertion in MLLMs

arXiv:2605.06115v2 Announce Type: replace Abstract: Multimodal Large Language Models (MLLMs), trained primarily on English-centric data, frequently generate culturally inappropriate or misaligned resp

CSR: Infinite-Horizon Real-Time Policies with Massive Cached State Representations

arXiv:2605.07325v1 Announce Type: cross Abstract: Deploying massive large language models (LLMs) as continuous cognitive engines for robotics is bottlenecked by the time-to-first-token (TTFT) latency

Decoding the Pulse of Reasoning VLMs in Multi-Image Understanding Tasks

arXiv:2603.04676v2 Announce Type: replace-cross Abstract: Multi-image reasoning remains a significant challenge for vision-language models (VLMs). We investigate a previously overlooked phenomenon: du

Divide and Conquer: Object Co-occurrence Helps Mitigate Simplicity Bias in OOD Detection

arXiv:2605.07821v1 Announce Type: cross Abstract: Out-of-distribution (OOD) detection is crucial for ensuring the reliability of deep learning models. Existing methods mostly focus on regular entangle

EgoPro-Bench: Benchmarking Personalized Proactive Interaction in Egocentric Video Streams

arXiv:2605.07299v1 Announce Type: cross Abstract: Existing Multimodal Large Language Models (MLLMs) remain primarily reactive, failing to continuously perceive environments or proactively assist users

Extracting Search Trees from LLM Reasoning Traces Reveals Myopic Planning

arXiv:2605.06840v1 Announce Type: new Abstract: Large language models (LLMs), especially reasoning models, generate extended chain-of-thought (CoT) reasoning that often contains explicit deliberation

researcharxiv-cs-ai

FAME: Forecasting Academic Impact via Continuous-Time Manifold Evolution

arXiv:2605.07208v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly used to brainstorm and evaluate research ideas, yet assessing such judgments is fundamentally difficult be

Gradient Extrapolation-Based Policy Optimization

arXiv:2605.06755v1 Announce Type: cross Abstract: Reinforcement learning is widely used to improve the reasoning ability of large language models, especially when answers can be automatically checked.

Is Your Prompt Poisoning Code? Defect Induction Rates and Security Mitigation Strategies

arXiv:2510.22944v2 Announce Type: replace-cross Abstract: Large language models (LLMs) have become indispensable for automated code generation, yet the quality and security of their outputs remain a c

MedAction: Towards Active Multi-turn Clinical Diagnostic LLMs

arXiv:2605.07305v1 Announce Type: cross Abstract: Most existing LLM diagnoses are evaluated on static, single-turn settings where complete patient information is provided upfront, an oversimplificatio

MiniAppBench: Evaluating the Shift from Text to Interactive HTML Responses in LLM-Powered Assistants

arXiv:2603.09652v3 Announce Type: replace Abstract: With the rapid advancement of Large Language Models (LLMs) in code generation, human-AI interaction is evolving from static text responses to dynami

MobileDev-Bench: A Benchmark for Issue Resolution in Mobile Application Development

arXiv:2603.24946v2 Announce Type: replace-cross Abstract: Large language models (LLMs) have shown strong performance on automated software engineering tasks, yet existing benchmarks focus primarily on

Multi-environment Invariance Learning with Missing Data

arXiv:2601.07247v2 Announce Type: replace-cross Abstract: Learning models that can handle distribution shifts is a key challenge in domain generalization. Invariance learning, an approach that focuses

safetyarxiv-cs-lg

OpenAI launches DeployCo to help businesses build around intelligence

OpenAI launched DeployCo, a new service designed to assist businesses in building and deploying applications leveraging OpenAI's AI models and intelligence capabilities. The offering appears to focus

model-releasesopenai