Search: “open-weights”

Beyond Symmetric Agents: Cognitive Diversity and Multi-Agent Debate in Small Language Models

arXiv:2609.35875v1 Announce Type: new Abstract: Multi-agent debate (MAD) reportedly improves reasoning and factuality over single-model inference, but prior work treats agents as symmetric peers, leav

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Model Releases

CompOrca: Corpus-Scale Compliance Labelling of Instruction-Tuning Data

DGX agent

arXiv:2609.37807v1 Announce Type: new Abstract: Studying how fine-tuning shapes refusal and noncompliance behaviour requires identifying training examples that refuse, evade or otherwise fail to fulfi

model-releasesarxiv-cs-cl

30 Sept 2026

CounterSteer: Suppressing Indirect Prompt Injection with Activation Steering

arXiv:2609.36570v1 Announce Type: cross Abstract: Indirect prompt injection makes an LLM agent treat untrusted retrieved text as instructions. We present CounterSteer, an inference-time defense that s

Evidence-Guided Schema Normalization for Temporal Tabular Reasoning

arXiv:2512.00329v2 Announce Type: replace-cross Abstract: Temporal reasoning over evolving semi-structured tables poses a challenge to current QA systems. We propose an approach that recasts the task

How Language Models Differ in Redistributing Attention-Head Activity Under Serial Demand

arXiv:2609.36221v1 Announce Type: new Abstract: The way a model distributes activity over each layer's attention heads offers a coarse view of how it routes information through depth; how this changes

It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs

arXiv:2609.37891v1 Announce Type: new Abstract: Current pre-training datasets are derived from web crawls, with all their issues, and were not designed to support mid- and post-training pipelines--for

researcharxiv-cs-cl

Know Thyself, Teach Thyself: Internal Information Flow for Selective Self-Distillation

arXiv:2609.36695v1 Announce Type: new Abstract: Self-distillation turns knowledge distillation into a closed learning loop and offers a path toward recursive self-improvement. Without an external teac

safetyarxiv-cs-lg

Language Models Are 'Insecure' Reporters

arXiv:2609.36139v1 Announce Type: cross Abstract: As large language models are deployed in increasingly autonomous long-horizon tasks, manually auditing and verifying the actions, artifacts, and outpu

Tutorials

LLMs learn different forms of metacognition when trained to predict their own accuracy

arXiv:2609.33886v2 Announce Type: replace Abstract: Large language models are trained to always produce an answer, regardless of whether they possess the relevant knowledge, which leads them to fabric

tutorialsarxiv-cs-cl

Lost in Translation: Measuring the Effect of Non-Native English on End User Performance of Large Language Models

arXiv:2609.36214v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used by people whose first language is not English, yet these users have been shown to receive systematica

researcharxiv-cs-cl

Hardware

Mixture-of-Kittens: MoE Megakernel for NVL72s

arXiv:2609.36070v1 Announce Type: cross Abstract: AI accelerator systems are rapidly consolidating into scale-up architectures, where tens to thousands of GPUs communicate over high-bandwidth, single-

hardwarearxiv-cs-lg

Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression

arXiv:2609.36322v1 Announce Type: cross Abstract: Chunked KV-cache compression reduces the memory and attention costs of long-context inference by compressing windows of consecutive tokens into fewer

PrivacySkills: How Privacy Guidance Shapes Source Selection in LLM Agents

arXiv:2609.35937v1 Announce Type: cross Abstract: While prior work has documented privacy failures in LLM agents, it remains unclear how the presentation of privacy guidance influences their choice of

researcharxiv-cs-ai

Prompted Identity Degrades Cooperation in Multi-Agent LLM Systems

arXiv:2609.35928v1 Announce Type: cross Abstract: Multi-agent LLM systems increasingly mix models from several providers, yet exposing each agent's underlying model identity to its peers significantly

RoboChrono: A Real Robot Benchmark for Streaming Task Understanding

model-releasesarxiv-cs-ro

arXiv:2609.36605v1 Announce Type: new Abstract: Understanding ongoing robot manipulation requires models to interpret visual observations in relation to interaction history and task progress. We intro

Same Bytes, Different Authority: Reserved-Token Representations in Chat-Template Prompt Injection

arXiv:2609.35932v1 Announce Type: cross Abstract: Prompt injection against LLM agents becomes much stronger when the injected instruction is wrapped in the model's own chat template. A forged template

Agents

Speed in the Blind Spot: An Interpretability Analysis of Dynamic Perception in VLMs for Autonomous Driving

arXiv:2609.37046v1 Announce Type: new Abstract: Vision-Language Models are increasingly used in autonomous-driving systems, yet their ability to recover dynamic physical state from visual input remain

agentsarxiv-cs-cv

The Surge of Anti-Semitism in German Social Media following the October 7 Attacks

arXiv:2609.36290v1 Announce Type: cross Abstract: We investigate the extent to which the Hamas attacks on Israel of October 7, 2023, have affected German social media debates about Judaism and Israel.

researcharxiv-cs-cl

When Models Don't Manipulate Manifolds: The Geometry of a Comparison Task

arXiv:2609.37680v1 Announce Type: cross Abstract: One of the current premises of mechanistic interpretability research is that detailed accounts of the geometry of neural network representations can t

When Should LLMs Trust Their Own Revisions? A Risk-Aware Study of Intrinsic Self-Correction

arXiv:2609.35832v1 Announce Type: cross Abstract: Intrinsic self-correction asks a language model to revise its own answer without receiving new external evidence. A second pass can recover mistakes,

Your Benchmark Is Not Saturated: Reviving Multiple-Choice Evaluation with Answer Pooling

model-releasesarxiv-cs-cl

arXiv:2609.37494v1 Announce Type: new Abstract: Multiple-choice benchmarks are cheap to grade and are running out of room, and the standard remedy, writing harder items, is slow and repeated for every

A Cheap Verifier is Good Enough: LLM Post-training is Robust to Erroneous Rewards

arXiv:2609.33467v1 Announce Type: new Abstract: When post-training large language models on tasks with semi-verifiable rewards, there are many factors (training steps, base model size, training order,

AutoRef: Harness Optimization for Agentic Multi-Reference Image Generation

arXiv:2609.35530v1 Announce Type: new Abstract: Recent image generation models can take multiple reference images as input and combine them into a new image. However, multi-reference image generation

Applications

AutoSDT: Scaling Data-Driven Discovery Tasks Toward Open Co-Scientists

arXiv:2506.08140v2 Announce Type: replace Abstract: Despite long-standing efforts in accelerating scientific discovery with AI, building AI co-scientists remains challenging due to limited high-qualit

applicationsarxiv-cs-lg

Beyond Token Savings: A Systematic Study of Context Compression in LLM Agents

arXiv:2609.32961v1 Announce Type: new Abstract: As LLM agents tackle longer tasks, they increasingly compress growing histories of reasoning, actions, and tool outputs. Compression can reduce token us

BudgetVerify: Budget-Tiered Verification for Financial QA

arXiv:2609.33052v1 Announce Type: cross Abstract: Financial question answering often requires precise numerical extraction, unit handling, and arithmetic over tables and text, but applying expensive v

researcharxiv-cs-lg

EngIntervene: Benchmarking Multimodal Engineering State Understanding and Design Intervention Reasoning

arXiv:2609.33261v1 Announce Type: new Abstract: Multimodal engineering benchmarks largely evaluate static understanding, such as recognizing components, interpreting diagrams, or answering technical q

Hardware

Greenpixie's AI Token Methodology: Assessing the Energy, Water and CO_2ext{-eq} Impact of AI Tokens for Open and Closed Weight Models

arXiv:2609.33965v1 Announce Type: cross Abstract: We describe a methodology for estimating the per-token energy cost of cloud-hosted large language model (LLM) inference, separating between input (pre

hardwarearxiv-cs-lg

InfiMed2: A Generalist Medical Multimodal Foundation Model from Contextual Evidence and Stability-Aware Supervision

arXiv:2609.34798v1 Announce Type: cross Abstract: Recent medical multimodal models have benefited from larger corpora, broader modality coverage, and stronger reasoning-oriented training, yet effectiv

researcharxiv-cs-cv

Nudgeability: Reasoning Models Follow Confidence Signals Without Tracking Their Own Competence

arXiv:2609.34572v1 Announce Type: cross Abstract: Reasoning language models that can call tools must decide during inference whether to answer unaided or delegate. Any self-reflection mechanism for th

PalmLeaf-VQA: A Multi-Script Visual Question Answering Benchmark for Historical Palm-Leaf Manuscript Understanding Across Diverse Regions

arXiv:2609.31651v1 Announce Type: new Abstract: Historical manuscripts remain largely absent from modern vision-language benchmarks, leaving open how well multimodal large language models (MLLMs) hand

Probe to Act: Elevating Browser-Use Agent via Active Visual Probing

arXiv:2609.33646v1 Announce Type: cross Abstract: Browser-use agents require seamless alignment between structured web metadata and visual information, while preserving relevant context across long in

Quantifying Behavioral Tails in Black-Box Language Models

arXiv:2609.33638v1 Announce Type: new Abstract: We introduce RareTrap, a framework for estimating the probability of severe behaviors in black box large language models (LLMs). A key challenge for pro

Re-derivability Decides What a Staged Agent Pipeline Recovers After an Upstream Fault

arXiv:2609.32802v1 Announce Type: cross Abstract: One variable sets what an upstream fault costs a staged pipeline of language-model agents: re-derivability, how much of what a stage needs it can rebu

Applications

Reading Too Much into Context: Passive Exposure Can Steer LLM Decisions

arXiv:2609.33065v1 Announce Type: cross Abstract: Large language model (LLM) assistants can now search the web and consult external sources while completing user requests. These sources can provide us

applicationsarxiv-cs-lg

SAGE: Symbolic Action-Gating and Editing for LLM Task Planners

model-releasesarxiv-cs-ro

arXiv:2609.34268v1 Announce Type: new Abstract: Large language models (LLMs) are now the default cognitive core of embodied household agents, yet the plans they emit are rarely checked against a groun

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

arXiv:2609.34970v1 Announce Type: new Abstract: Safety-aligned LLMs can exhibit emergent misalignment (EM): narrow domain adaptation unexpectedly triggers catastrophic safety failures across unrelated

When Do Model Internals Help? Exploring the Role of Representation Engineering in LLM Safety

arXiv:2609.34771v1 Announce Type: cross Abstract: Reliable AI safeguards require both control mechanisms that reduce unsafe behavior and monitoring mechanisms that detect safety risks during model int

safetyarxiv-cs-lg

AcoustiClaim: A Numeric Claim Benchmark with Instrument Ground Truth

model-releasesarxiv-cs-cl

arXiv:2609.30483v1 Announce Type: cross Abstract: Audio language models state numbers for acoustic quantities, and neither human opinion nor a judge model says whether such a number is true of the sig

Local Ai

Cheap, open agents make LLM pollution harder to mitigate

arXiv:2609.31054v1 Announce Type: new Abstract: Large Language Model (LLM) pollution occurs when synthetic responses contaminate data intended to capture human behavior. High deployment costs have so

local-aiarxiv-cs-ai

Multi-agent Scaling Across Disjunctive and Compensatory Tasks

arXiv:2609.31563v1 Announce Type: new Abstract: Multi-agent LLM systems are often expected to improve as team size increases, yet the scaling behavior may depend on task structure. Our central contrib

safetyarxiv-cs-ai

Muslim: A Deployed Arabic Voice AI Platform for Grounded Islamic Knowledge

model-releasesarxiv-cs-cl

arXiv:2609.31511v1 Announce Type: new Abstract: We present Muslim, a production Arabic voice AI platform serving grounded, sourced Islamic knowledge to real users. Beyond a real-time voice pipeline (N

Privacy-Preserving Prompted Policy Search for Robotic Control

arXiv:2609.30554v1 Announce Type: new Abstract: Large language models (LLMs) have recently demonstrated promising capabilities as in-context policy optimizers for Reinforcement Learning (RL), enabling

safetyarxiv-cs-ro

AgenticCADedit: A Stateful, Tool-Mediated Agentic Approach to Multimodal 3D CAD Editing

arXiv:2609.29621v1 Announce Type: new Abstract: Computer-aided design is central to industrial manufacturing, and much of a designer's daily work consists of editing existing models from multimodal re

Agents

An Empirical Study of VLM Pipelines for Long-Document QA

arXiv:2609.29933v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) are increasingly used for long-document processing, where the inputs combine text with charts, tables, figures, and comp

agentsarxiv-cs-ai

Agents

ASIRF: An Agentic Framework for Context-Dependent Sensitive Information Redaction

arXiv:2609.29191v1 Announce Type: new Abstract: Sensitive information is defined by domain and intent, not a universal category, yet redaction systems such as privacy filters and named-entity recogniz

agentsarxiv-cs-ai

Augur: A Synthetic Decision Lab for Rehearsing Reactions to Product and Policy Changes

arXiv:2609.29952v1 Announce Type: new Abstract: Before a product or policy change ships, the question that matters is how people will react to it. Augur rehearses that reaction offline: it builds a ty

safetyarxiv-cs-ai

Broadening Uncertainty Estimation for Audio Question Answering Across Methods, Formats, and Inputs

arXiv:2609.28879v1 Announce Type: cross Abstract: Audio-language models can produce confident answers unsupported by the audio, motivating uncertainty estimates that identify unreliable responses. We

researcharxiv-cs-ai