Search: “llm”

Sparse Autoencoders Are a Shared Coordinate System for Symbolic Matching

model-releasesarxiv-cs-cl

arXiv:2610.19636v1 Announce Type: cross Abstract: Large language models (LLMs) hold broad knowledge in one set of weights, but lack interpretability and trustworthy reasoning, especially in high-stake

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

SpecCloak: Probing Harmful Code Generation through Engineering Specifications

DGX agent

arXiv:2610.17907v1 Announce Type: cross Abstract: The code generation capabilities of large language models (LLMs) can equally be misused to implement harmful functionality, underscoring the need to u

safetyarxiv-cs-ai

15 Oct 2026

SpecFit: Not All Draft Tokens Cost the Same in Offloaded MoE Inference

arXiv:2610.15417v1 Announce Type: new Abstract: Mixture-of-experts (MoE) models activate only a few experts per token, but when their expert weights are offloaded to host memory, host-to-device (H2D)

researcharxiv-cs-ai

Spectral-MoE: Heterogeneous Per-Layer Expert Utilization in Dense-to-MoE Conversion

arXiv:2610.19232v1 Announce Type: new Abstract: Large language models impose substantial memory demands and arithmetic costs during inference. Dense-to-mixture-of-experts (MoE) conversion is a model o

The Reasoning Trap -- Logical Reasoning as a Mechanistic Pathway to Situational Awareness

model-releasesarxiv-cs-cl

arXiv:2603.09200v2 Announce Type: replace-cross Abstract: Situational awareness, the capacity of an AI system to recognize its own nature, understand its training and deployment context, and reason st

The Silent Failures in Agentic Data Science

arXiv:2610.14115v1 Announce Type: new Abstract: Large language models have significantly lowered the barrier to writing code. Naturally, this has also led to the development of data science agents tha

agentsarxiv-cs-ai

Think in Strategies, Act with Tools: Traceable Agentic Optimization for Combinatorial Problems

arXiv:2610.14736v1 Announce Type: new Abstract: Optimization by PROmpting (OPRO) directly leverages Large Language Models (LLMs) to generate candidate solutions through natural-language prompting. How

agentsarxiv-cs-ai

Thinking Outside the Bots: Human Ideas in AI's Semantic Blind Spots Are Not Inferior

arXiv:2610.17324v1 Announce Type: cross Abstract: AI-generated outputs are frequently perceived as highly creative, but also tend to be more similar to each other (homogenized) than human ideas. Does

researcharxiv-cs-ai

TIMERSI: Towards Self-Improved Agentic Deployment for Time-Series Foundation Models

arXiv:2610.19725v1 Announce Type: new Abstract: Time-series foundation models provide strong zero-shot forecast, but deployment for real- world applications still requires task-specific designs for mo

agentsarxiv-cs-lg

To Reason but Not to Stop: Understanding and Recovering Pruned Large Reasoning Models

arXiv:2610.19934v1 Announce Type: new Abstract: Weight pruning can degrade the accuracy of large reasoning models (LRMs), yet it remains unclear whether it damages reasoning itself. We find that much

researcharxiv-cs-cl

Toward System One Clinical Intelligence: Parameter-Efficient Fine-Tuning and Counterfactual Explainability of Decision Foundation Models

arXiv:2610.15029v1 Announce Type: cross Abstract: Automated disease classification from patient-reported symptoms requires balancing the speed of lightweight discriminative models against the contextu

Towards a Data-Centric Understanding of Spatial Reasoning in Multimodal Large Language Models

model-releasesarxiv-cs-cl

arXiv:2610.19108v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) have been widely applied in embodied agents and robotics for vision tasks. However, MLLMs suffer from limitat

UDHRBench: Benchmarking Human Rights in Large Language Models

arXiv:2610.15179v1 Announce Type: cross Abstract: Large language models are increasingly deployed in contexts requiring adherence to ethical and legal norms, yet their alignment with internationally r

Unlocking Token Rewards via Training-Free Reward Attribution

arXiv:2610.18704v1 Announce Type: new Abstract: In this paper, we propose an extremely efficient, training-free method to extract token-level reward signals directly from an existing deep reward model

VETNET: Assembling Benchmark Datasets for AI-Driven Veterinary Innovation

arXiv:2610.14149v1 Announce Type: new Abstract: VETNET represents an ecosystem of live benchmark datasets and evaluation frameworks designed to accelerate AI-driven innovation in animal science and ve

VQEL: Enabling Self-Play in Emergent Language Games via Agent-Internal Vector Quantization

arXiv:2503.04940v3 Announce Type: replace-cross Abstract: Emergent Language (EL) focuses on the emergence of communication among artificial agents. Although symbolic communication channels more closel

safetyarxiv-cs-ai

We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI. Post-training results…

model-releasesjeremy-howard--x

We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI. Post-training results coming soon -- stay tuned! Lots of details in our technical

WeFair: Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in Short-Video Search

arXiv:2610.16927v1 Announce Type: new Abstract: Short-video search requires compact query rewriters that infer personalized intent from brief, ambiguous queries under strict latency constraints. The c

safetyarxiv-cs-ai

Tutorials

What Does LALM Post-training Learn? Rethinking Audio Understanding and Reasoning Capabilities

arXiv:2610.19476v1 Announce Type: cross Abstract: Post-training often improves the performance of large audio-language models on audio understanding and reasoning benchmarks. However, these improvemen

tutorialsarxiv-cs-cl

What Is a Clarification? Estimating Model-Dependent Input Ambiguity

arXiv:2610.20052v1 Announce Type: new Abstract: Ambiguous inputs can admit multiple interpretations, each potentially yielding a different answer. Existing work largely treats this ambiguity as a prop

researcharxiv-cs-cl

When Does Momentum Help in Reinforcement Learning with Verifiable Rewards?

arXiv:2610.20413v1 Announce Type: new Abstract: Momentum is a classical tool for accelerating convex optimization and a standard component of modern large language model training. Yet recent empirical

When Non-Adversarial Noise Hurts Learning: Benchmarking Skill Self-Evolution in Agents

arXiv:2610.19910v1 Announce Type: new Abstract: Large language model agents increasingly improve by distilling experience into reusable skills, but errors in that experience can persist in learned rul

When Should Enterprise Policy Decisions Be Learned, and When Should They Be Reasoned?

arXiv:2610.17643v1 Announce Type: new Abstract: An enterprise agent that answers under a written policy must choose an action before it says anything: answer, ask for a missing fact, ask for confirmat

Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages

arXiv:2610.18420v1 Announce Type: cross Abstract: We trained Agaram 2.5B, a 2.5-billion-parameter Tamil-English language model, from scratch on about 50 billion tokens (58% Tamil, the rest English, ma

Applications

AIRRI: Towards a Human-Readable, Machine-Resistant Reading Interface

arXiv:2610.18724v1 Announce Type: new Abstract: The rapid advancement of image-to-text extraction systems and large language models (LLMs) poses new challenges for educational assessment by making ass

applicationsarxiv-cs-cv

Beyond Adjacent Layers: Graph-Guided Layer Fusion for Compressing Large Language Models

arXiv:2610.15159v1 Announce Type: new Abstract: Existing depth-compression methods for large language models (LLMs) score layers individually or group only adjacent or contiguous layers, implicitly as

Checking the Checkers: A Synthesis of Verifier Noise and Robustness in RLVR

arXiv:2610.17824v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a principal recipe for improving the reasoning of large language models (LLMs). It rest

ChemExplain: A Multimodal Benchmark for Explanatory Chemical Reasoning

model-releasesarxiv-cs-cv

arXiv:2610.17183v1 Announce Type: cross Abstract: Current chemistry benchmarks primarily evaluate outcome prediction or reaction pathway generation, but rarely test whether a model can explain why an

Contrastive Covariance Subspaces for Activation Steering

arXiv:2610.14025v1 Announce Type: new Abstract: Activation steering controls large language model behavior by adding a direction to hidden activations during inference. For steering to work well, that

researcharxiv-cs-lg

D-World: An Explicit World-Language-Action Model with Recursive Self-Improvement under Dual Control

model-releasesarxiv-cs-ro

arXiv:2610.17114v1 Announce Type: new Abstract: Frontier large language models (LLMs) now reason about space well enough to program robots, yet the world models behind robot policies are implicit, hel

DASS: A Solver-Adaptable Dynamic Auxiliary Search Strategy for Symbolic Regression

arXiv:2610.16812v1 Announce Type: new Abstract: Symbolic regression (SR) aims to discover interpretable mathematical expressions from data and plays a key role in scientific discovery. However, existi

researcharxiv-cs-lg

Density-Matrix Reinforcement Learning

arXiv:2610.17214v1 Announce Type: new Abstract: What information does a reinforcement learning policy omit? At each state, a standard policy stores one probability per action. It has no place to recor

DVTR-KD: Distilling Vision-Language Models via Intermediate Visual Token Reconstruction

arXiv:2610.17853v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) have achieved strong performance on diverse vision-language tasks, but their computational demands limit deploy

researcharxiv-cs-cv

Exploring Jailbreak Vulnerability Across Depth-Based Dynamic Inference in Large Language Models

arXiv:2610.15207v1 Announce Type: new Abstract: Large Language Models (LLMs) have been shown to be computationally expensive and vulnerable to jailbreak attacks, in which adversarial prompts bypass sa

Hardware

FastEEP: Fast Intra-Instance Elastic Scaling for Expert-Parallel MoE Serving

arXiv:2610.18451v1 Announce Type: cross Abstract: Large-scale expert parallelism (EP) has become the standard deployment configuration for Mixture-of-Experts (MoE) models in production. Meanwhile, the

hardwarearxiv-cs-lg

Local Ai

Federate the Router: Learning Language Model Routers with Sparse and Decentralized Evaluations

arXiv:2601.22318v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly accessed as remotely hosted services by edge and enterprise clients that cannot run frontier models lo

local-aiarxiv-cs-lg

FoveaTree: Information-Density Trees for Query-Guided Visual Token Pruning in Video Large Language Models

model-releasesarxiv-cs-cv

arXiv:2610.16744v1 Announce Type: new Abstract: Video large language models (Video LLMs) encode each frame into hundreds of visual tokens, making video inference expensive. Existing methods reduce thi

Applications

Generalizable Model Editing Could Be More Than Direct Generalization

arXiv:2610.15707v1 Announce Type: new Abstract: A central desideratum in model editing is generalization to complex tasks, which prior work formalizes under a direct generalization paradigm: treating

applicationsarxiv-cs-lg

GIFT-Eval-CTX: A Benchmark For Contextual Forecasting Model Evaluation

arXiv:2610.17794v1 Announce Type: new Abstract: Contextual forecasting seeks to improve time series forecasting by incorporating forward-looking information, such as planned events, external condition

Tutorials

Gradient Routing Localizes and Removes Unintended Behaviors in RL

arXiv:2610.17062v1 Announce Type: new Abstract: We propose a mitigation to reward hacking in reinforcement learning. Our setup involves a misspecified reward function that gives rise to unintended beh

tutorialsarxiv-cs-lg

Growing to See: Efficient Pretraining of Self-Supervised Vision Transformers

arXiv:2610.16817v1 Announce Type: new Abstract: Progressive depth growth builds large networks by starting with shallow models and inserting layers during training. Recent advances in large language m

researcharxiv-cs-cv

GUARD: Guided Unlearning and Retention via Data Attribution for Large Language Models

arXiv:2506.10946v3 Announce Type: replace Abstract: Unlearning large language models (LLMs) is important due to copyright and privacy concerns. A key challenge is unintended forgetting, where removing

HelpMeOut: A Neuropsychology-Grounded Benchmark for Error-Aware Monitoring of Household Task Execution

model-releasesarxiv-cs-ro

arXiv:2610.17694v1 Announce Type: new Abstract: We introduce HelpMeOut, an embodied AI benchmark for error-aware monitoring of household task execution. Existing embodied household benchmarks typicall

IMAGAgent: Orchestrating Multi-Turn Image Editing via Constraint-Aware Planning and Reflection

arXiv:2603.29602v2 Announce Type: replace-cross Abstract: Existing multi-turn image editing paradigms are often confined to isolated single-step execution. Due to a lack of context-awareness and close

agentsarxiv-cs-cv

Learning Temporal Evidence Selection for Video Anomaly Understanding

arXiv:2610.16299v1 Announce Type: new Abstract: Video anomaly understanding (VAU) aims to generate language-level interpretations of anomalous events in long videos, requiring visual evidence that sup

researcharxiv-cs-cv

MARL-BattleGrounds: A Novel JAX-native Benchmark for Heterogeneous and Competitive Multi-agent RL

arXiv:2610.14883v1 Announce Type: cross Abstract: We present MARL-BattleGrounds (MARL-BGs), an easy-to-use JAX-native benchmark for heterogeneous and competitive multi-agent reinforcement learning (MA

mathbf{S^2LM}: Towards Semantic Steganography via Large Language Models

model-releasesarxiv-cs-cv

arXiv:2511.05319v3 Announce Type: replace Abstract: Despite remarkable progress in steganography, embedding sentence-level information into cover media remains challenging. In this work, we present th

Local Ai

microsoft/UserRepGen-4B

https://preview.redd.it/gwb4of5vmivh1.png?width=2400&format=png&auto=webp&s=853f4cf25b4140ccf8b151b4124547788a544b33 UserRepGen-4B is a fine-tuned Qwen3.5-4B language model that generates a structured

local-air-localllama