Search: “llm”

The Scaling Laws of Skills in LLM Agent Systems

arXiv:2605.16508v1 Announce Type: cross Abstract: As agent systems scale, skills accumulate into large reusable libraries, yet their scaling laws remain poorly understood. Across 15 frontier LLMs, 1,1

19 May 2026

Hardware

VeriCache: Turning Lossy KV Cache into Lossless LLM Inference

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

DGX agent

arXiv:2605.17613v1 Announce Type: cross Abstract: The large size of the KV cache has become a major bottleneck for serving LLMs with increasing context lengths. In response, many KV cache compression

hardwarearxiv-cs-lg

19 May 2026

A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning

arXiv:2510.18814v3 Announce Type: replace-cross Abstract: Can language models improve their reasoning performance without external rewards, using only their own sampled responses for training? We show

Graph-Regularized Sparse Autoencoders for LLM Safety Steering

arXiv:2512.06655v3 Announce Type: replace-cross Abstract: Sparse autoencoders (SAEs) are increasingly used to extract activation directions for inference-time steering, but their standard sparsity obj

Look Before You Leap: Autonomous Exploration for LLM Agents

arXiv:2605.16143v1 Announce Type: new Abstract: Large language model based agents often fail in unfamiliar environments due to premature exploitation: a tendency to act on prior knowledge before acqui

agentsarxiv-cs-ai

LPDS: Evaluating LLM Robustness Through Logic-Preserving Difficulty Scaling

arXiv:2605.15393v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly deployed to perform tasks with minimal human oversight, it is crucial that these models operate robustl

researcharxiv-cs-lg

Optimizing LLM Inference: Fluid-Guided Online Scheduling with Memory Constraints

arXiv:2504.11320v3 Announce Type: replace-cross Abstract: Large language models now serve millions of users daily, with providers incurring costs exceeding $700,000 per day. Each request requires toke

RecMem: Recurrence-based Memory Consolidation for Efficient and Effective Long-Running LLM Agents

arXiv:2605.16045v1 Announce Type: cross Abstract: Memory systems often organize user-agent interactions as retrievable external memory and are crucial for long-running agents by overcoming the limited

agentsarxiv-cs-ai

Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation

arXiv:2605.15239v1 Announce Type: new Abstract: Safety alignment often improves robustness to harmful queries at the cost of reasoning ability, a tradeoff known as the safety tax. A common cause is di

SafeGPT: Preventing Data Leakage and Unethical Outputs in Enterprise LLM Use

arXiv:2601.06366v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) are transforming enterprise workflows but introduce security and ethics challenges when employees inadvertently s

Structure-BiEval: A Self-Supervised, Dual-Track Framework for Decoupling Structure and Content in LLM Evaluation for Web Information Systems

arXiv:2601.19923v2 Announce Type: replace-cross Abstract: As Large Language Models (LLMs) evolve into the core of Web-based autonomous agents and complex Web Information Systems, their ability to fait

TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning

arXiv:2505.15692v5 Announce Type: replace Abstract: Reinforcement learning (RL) has emerged as an effective paradigm for enhancing model reasoning. However, existing RL methods like GRPO typically rel

safetyarxiv-cs-cl

A Heterogeneous Temporal Memory Governance Framework for Long-Term LLM Persona Consistency

arXiv:2605.14802v1 Announce Type: new Abstract: Large language models often suffer from fact loss, timeline confusion, persona drift, and reduced stability during long-range interaction, especially un

Boosting LLM Reasoning via Human-Inspired Reward Shaping

arXiv:2602.04265v3 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a promising paradigm for enhancing reasoning in Large Language Models (LL

Case-Based Calibration of Adaptive Reasoning and Execution for LLM Tool Use

arXiv:2605.15041v1 Announce Type: new Abstract: Tool use extends large language models beyond parametric knowledge, but reliable execution requires balancing appropriate reasoning depth with strict st

agentsarxiv-cs-ai

Invisible Orchestrators Suppress Protective Behavior and Dissociate Power-Holders: Safety Risks in Multi-Agent LLM Systems

arXiv:2605.13851v1 Announce Type: new Abstract: Multi-agent orchestration -- in which a hidden coordinator manages specialized worker agents -- is becoming the default architecture for enterprise AI d

LLM-based Detection of Manipulative Political Narratives

arXiv:2605.14354v1 Announce Type: new Abstract: We present a new computational framework for detecting and structuring manipulative political narratives. A task that became more important due to the s

researcharxiv-cs-cl

LLM-Based Robustness Testing of Microservice Applications: An Empirical Study

arXiv:2605.14202v1 Announce Type: cross Abstract: Malformed, missing, or boundary-value inputs in microservice APIs can cascade across dependent services, threatening reliability. Robustness testing s

MUON+: Towards More Effective Muon via One Additional Normalization Step for LLM Pre-training

arXiv:2602.21545v3 Announce Type: replace Abstract: Muon has recently emerged as a strong optimizer for large language model pre-training, orthogonalizing the momentum matrix via Newton--Schulz polar

Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks

arXiv:2605.14604v1 Announce Type: new Abstract: This position paper argues that effective tutoring requires corrective friction: surfacing misconceptions and challenging them supportively to drive con

Data Difficulty and the Generalization--Extrapolation Tradeoff in LLM Fine-Tuning

arXiv:2605.12906v1 Announce Type: cross Abstract: Data selection during supervised fine-tuning (SFT) can critically change the behavior of large language models (LLMs). Although existing work has stud

FOAM: Blocked State Folding for Memory-Efficient LLM Training

arXiv:2512.07112v2 Announce Type: replace-cross Abstract: Large language models (LLMs) have demonstrated remarkable performance due to their large parameter counts and extensive training data. However

History Anchors: How Prior Behavior Steers LLM Decisions Toward Unsafe Actions

arXiv:2605.13825v1 Announce Type: new Abstract: Frontier LLMs are increasingly deployed as agents that pick the next action after a long log of prior tool calls produced by the same or a different mod

Is Video Anomaly Detection Misframed? Evidence from LLM-Based and Multi-Scene Models

arXiv:2605.12725v1 Announce Type: new Abstract: Recent video anomaly detection research has expanded rapidly with an emphasis on general models of normality intended to work across many different scen

safetyarxiv-cs-cv

Applications

Learning Perturbations to Extrapolate Your LLM

arXiv:2605.13284v1 Announce Type: cross Abstract: Recent advancements in large language models demonstrate that injecting perturbations can substantially enhance extrapolation performance. However, cu

applicationsarxiv-cs-lg

REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations

arXiv:2605.12813v1 Announce Type: cross Abstract: Large language models (LLMs) achieve strong performance across many tasks but remain vulnerable to hallucinations, motivating the need for realistic a

Respecting Self-Uncertainty in On-Policy Self-Distillation for Efficient LLM Reasoning

arXiv:2605.13255v1 Announce Type: new Abstract: On-policy self-distillation trains a reasoning model on its own rollouts while a teacher, often the same model conditioned on privileged context, provid

Revisiting DAgger in the Era of LLM-Agents

arXiv:2605.12913v1 Announce Type: new Abstract: Long-horizon LM agents learn from multi-turn interaction, where a single early mistake can alter the subsequent state distribution and derail the whole

Simulating Students or Sycophantic Problem Solving? On Misconception Faithfulness of LLM Simulators

arXiv:2605.12748v1 Announce Type: cross Abstract: Large language models (LLMs) can fluently generate student-like responses, making them attractive as simulated students for training and evaluating AI

Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoning

arXiv:2602.06475v2 Announce Type: replace Abstract: Large language models (LLMs) excel at complex tasks with advances in reasoning capabilities. However, existing reward mechanisms remain tightly coup

Tracing Persona Vectors Through LLM Pretraining

arXiv:2605.13329v1 Announce Type: cross Abstract: How large language models internally represent high-level behaviors is a core interpretability question with direct relevance to AI safety: it determi

B4DL: A Benchmark for 4D LiDAR LLM in Spatio-Temporal Understanding

model-releasesarxiv-cs-cv

arXiv:2508.05269v2 Announce Type: replace Abstract: Understanding dynamic outdoor environments requires capturing complex object interactions and their evolution over time. LiDAR-based 4D point clouds

CATS: Cascaded Adaptive Tree Speculation for Memory-Limited LLM Inference Acceleration

arXiv:2605.11186v1 Announce Type: new Abstract: Auto-regressive decoding in Large Language Models (LLMs) is inherently memory-bound: every generation step requires loading the model weights and interm

Efficient LLM Reasoning via Variational Posterior Guidance with Efficiency Awareness

arXiv:2605.11019v1 Announce Type: new Abstract: Although large language models rely on chain-of-thought for complex reasoning, the overthinking phenomenon severely degrades inference efficiency. Exist

Enabling Performant and Flexible Model-Internal Observability for LLM Inference

arXiv:2605.11093v1 Announce Type: new Abstract: Today's inference-time workloads increasingly depend on timely access to a model's internal states. We present DMI-Lib, a high-speed deep model inspecto

OGLS-SD: On-Policy Self-Distillation with Outcome-Guided Logit Steering for LLM Reasoning

arXiv:2605.12400v1 Announce Type: new Abstract: We study {on-policy self-distillation} (OPSD), where a language model improves its reasoning ability by distilling privileged teacher distributions alon

Local Ai

SOMA: Efficient Multi-turn LLM Serving via Small Language Model

arXiv:2605.11317v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly deployed in multi-turn dialogue settings where preserving conversational context across turns is essential

local-aiarxiv-cs-cl

Targeted Tests for LLM Reasoning: An Audit-Constrained Protocol

arXiv:2605.11599v1 Announce Type: new Abstract: Fixed reasoning benchmarks evaluate canonical prompts, but semantically valid changes in presentation can still change model behavior. Studies of prompt

researcharxiv-cs-lg

Local Ai

TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection

arXiv:2605.12456v1 Announce Type: cross Abstract: We introduce TextSeal, a state-of-the-art watermark for large language models. Building on Gumbel-max sampling, TextSeal introduces dual-key generatio

local-aiarxiv-cs-cl

The Algorithmic Caricature: Auditing LLM-Generated Political Discourse Across Crisis Events

arXiv:2605.12452v1 Announce Type: new Abstract: Large Language Models (LLMs) can generate fluent political text at scale, raising concerns about synthetic discourse during crises and social conflict.

researcharxiv-cs-cl

Unlocking LLM Creativity in Science through Analogical Reasoning

arXiv:2605.11258v1 Announce Type: cross Abstract: Autonomous science promises to augment scientific discovery, particularly in complex fields like biomedicine. However, this requires AI systems that c

agentsarxiv-cs-cl

Aligning LLM Uncertainty with Human Disagreement in Subjectivity Analysis

arXiv:2605.10415v1 Announce Type: new Abstract: Large language models for subjectivity analysis are typically trained with aggregated labels, which compress variations in human judgment into a single

safetyarxiv-cs-cl

Beyond Accuracy: Evaluating Strategy Diversity in LLM Mathematical Reasoning

arXiv:2605.09292v1 Announce Type: new Abstract: Large language models now achieve high final-answer accuracy on mathematical reasoning benchmarks, but accuracy alone does not capture reasoning flexibi

Beyond Red-Teaming: Formal Guarantees of LLM Guardrail Classifiers

arXiv:2605.10901v1 Announce Type: new Abstract: Guardrail Classifiers defend production language models against harmful behavior, but although results seem promising in testing, they provide no formal

BiAxisAudit: A Novel Framework to Evaluate LLM Bias Across Prompt Sensitivity and Response-Layer Divergence

model-releasesarxiv-cs-cl

arXiv:2605.09041v1 Announce Type: new Abstract: Bias audits of large language models now operate within governance frameworks such as the EU AI Act, making benchmark reliability a security concern in

Black-Box Detection of LLM-Generated Text Using Generalized Jensen-Shannon Divergence

arXiv:2510.07500v2 Announce Type: replace Abstract: We study black-box detection of machine-generated text under practical constraints: the scoring model (proxy LM) may mismatch the unknown source mod

researcharxiv-cs-lg

Curriculum Learning for LLM Pretraining: An Analysis of Learning Dynamics

arXiv:2601.21698v2 Announce Type: replace-cross Abstract: Curriculum learning changes the order of pretraining data, but it remains unclear how ordering changes the learning dynamics. We pretrain mode

DiagnosticIQ: A Benchmark for LLM-Based Industrial Maintenance Action Recommendation from Symbolic Rules

arXiv:2605.08614v1 Announce Type: new Abstract: Monitoring complex industrial assets relies on engineer-authored symbolic rules that trigger based on sensor conditions and prompt technicians to perfor