Live intelligence

Latest AI updates

27 updates

Research

AttnSR: Attention Step Reward for Long-Horizon Agentic RL

AttnSR is a step-level reward augmentation method for long-horizon LLM agents that derives reward allocations directly from policy attention during training replay, eliminating the need for a separate reward model or additional rollouts. By tracing backward through the policy model's attention graph, it estimates the relevance of earlier steps to the final response and uses a running baseline to determine whether rewards reinforce or penalize selected steps. In experiments with Qwen2.5-7B-Instruct, AttnSR improved ALFWorld success from 91.2% to 97.1% and consistently outperformed PPO on ScienceWorld while reducing environment interactions by 29.9%.

arxiv15h agoScore 75

reinforcement learningLLM agentsreward shaping

Research

tok: A Textual Concrete Syntax for BPMN/DMN with LLM and MCP Support

The paper presents tok, a plain-text notation for BPMN 2.0.2 process models and DMN 1.5 decisions, with an engine that checks documents against both standards, compiles them to XML with automatic diagrams, and serves language-model applications through the Model Context Protocol. It reports that tok models use fewer tokens than BPMN XML, that its checker detects 98.2% of injected faults versus 85.8% for XML Schema with bpmnlint, and that new language models write valid tok from a one-page reference. Agent evaluations with a lean skill and checker tool reached 97.6% to 100% valid tok across 87 process descriptions, and 44 of 47 simulated event logs fit their source models.

arxiv15h agoScore 73

BPMNDMNModel Context Protocol

Research

ChemE-as-Code: An Open-Source Platform for Agentic AI in Chemical Engineering

Researchers introduce ChemE-as-Code, an open-source environment where LLM-based agents can perform chemical engineering tasks using calculation tools spanning catalysis, process simulation, plant layout, HAZOP and life cycle assessment. Engineering files are machine-readable, versioned in Git, and calculations run outside the LLM, with each value recording its origin and automated checks flagging inconsistencies and out-of-date results by file and line. In an illustrative ammonia case, agents produced a preliminary basic-engineering package for USD 224 of API cost; planted faults and engineering changes showed the checks catch only what their rules cover, so they support rather than replace human review. The design has not been reviewed by an independent engineer.

arxiv15h agoScore 69

agentic AIchemical engineeringLLM agents

Research

Harness Design Alters Fixed LLM Compliance With Banking Conduct Rules

A new arXiv study holds four open-weight models fixed while running each through five agent harnesses on 156 pre-registered simulated retail-banking cases, scoring 3,120 two-turn episodes from executed tool traces. Pooled correct-case rates ranged only from 87.3% to 92.1%, but within one model the best-to-worst harness spread was 11.5 percentage points, enough to change its rank from first to last. A same-model compliance reviewer cut violations from 9.6% to 6.4% without increasing over-refusals, while rules requiring date arithmetic failed in every harness. In an exploratory analysis, 26% of violations included a customer message denying the action the agent had just executed, which transcript-only review would miss.

arxiv1d agoScore 74

AI governanceLLM agentsfinancial services compliance

Research Editor selected

HarnessFlow: MCTS-Guided Harness Optimization for LLM Agents

HarnessFlow is a framework that automatically optimizes the harness (the code governing tool interaction) for LLM agents under a limited evaluation budget. It uses a Monte Carlo Tree Search policy to balance exploration and exploitation, with a coding agent refining harnesses based on execution feedback. Across six benchmarks, it achieves an average score of 69.48%, outperforming initial harnesses by 10.89 points and Meta-Harness by 3.95 points, while reducing per-task execution costs.

Why it matters: This introduces a novel, practical method for automating harness design, showing significant performance gains and cost efficiency, which is valuable for AI agent development.

arxiv1d agoScore 85

LLM agentsharness optimizationMCTS

Research

LegalEvolver: Rubric-Guided Self-Evolution for Legal Judgment Prediction

LegalEvolver is a query optimization framework for legal judgment prediction that improves retrieval by generating legally informative queries. It uses heuristic seed queries and rubric-guided reinforcement learning to iteratively self-evolve, surpassing initial performance ceilings. Benchmarks show effectiveness over direct SFT+RL and fixed rubrics.

arxiv1d agoScore 78

Legal Judgment PredictionQuery OptimizationSelf-Evolution

Research Editor selected

Coca-Cola: Latent Context Compression for LLM Agents

Coca-Cola compresses tool observations into latent tokens embedded in the agent's input space, reducing token consumption by 70-89% and runtime by 24-56% while improving task performance on QA benchmarks. Action-Preserving Training aligns next-action distributions to retain decision-critical information.

Why it matters: Offers a concrete, novel method for agent context compression with significant efficiency gains and open-source code.

arxiv1d agoScore 88

LLM agentscontext compressionlatent representation

Research Editor selected

Online Rubric Self-Distillation for Token-Level Credit Assignment in RL

The paper introduces ORSD, a method that generates natural-language process rubrics from on-policy trajectories to provide dense token-level supervision for reinforcement learning. A frozen verifier selects the best rubric bundle based on alignment with task success, and the likelihood gap with and without the rubric yields token-specific credit added to trajectory-level advantages. Experiments on ALFWorld and WebShop show faster policy improvement and higher final performance for Qwen3-1.7B and Qwen3-4B.

Why it matters: This is a novel, concrete AI research contribution with empirical results, addressing a key challenge in RL credit assignment.

arxiv1d agoScore 85

reinforcement learningtoken-level credit assignmentprocess rubrics

Research

Turn-Level Best-of-N Sampling Boosts LLM Agent Performance

This paper introduces turn-level best-of-N sampling, a simple inference-time scaling method for LLM agents on long-horizon tasks. Using a generator-selector framework with two off-the-shelf open-weight LLMs, the approach improves accuracy by 5-10% and consistency across repeated attempts. Analysis shows agents adopt desirable behaviors like fewer attempts, better instruction following, and improved debugging.

arxiv1d agoScore 72

LLM agentsinference-time scalingsampling methods

Research

Lessons from Building an LLM Agent for Automated Experimentation

The paper reports practical lessons from building and operating an LLM-based agent for automating experimentation workflows. The central finding is that nondeterminism should be bounded by having the model select and sequence procedures while deterministic code computes all reported quantities. Surprising outcomes include improved answers when information is withheld, high demand for mundane retrieval tasks, and risks to statistical validity from the agent's eagerness to help.

arxiv2d agoScore 72

LLM agentsscientific experimentationreliability

Research Editor selected

GeoFaham: Multi-Agent Geospatial Analytics for Disaster Response

GeoFaham is a domain-specialized multi-agent system that decomposes geospatial disaster-response workflows into specialist agents for vector, map, imagery, and raster analysis, coordinated by an Orchestrator with a dual-ledger control loop. It uses a Response Store to persist outputs outside the LLM context, reducing token usage. On the new GeoFaham-Bench benchmark, it outperforms general coding agents with a score of 0.940 versus 0.879, using 7.5 times fewer tokens.

Why it matters: Introduces a novel multi-agent architecture and benchmark for geospatial AI, showing significant efficiency and accuracy gains over strong baselines.

arxiv2d agoScore 88

multi-agent systemsgeospatial analyticsdisaster response

Research Editor selected

LatentTriad: Pre-Action Risk Representations for Malicious Agent Skill Detection

LatentTriad is a new method that detects malicious skills in LLM agents without executing them, using hidden states from a single prefill pass. It combines contextual boundary features, group-adjusted operation directions, and departures from benign state relationships to achieve 99.0-99.2% pooled F1 across 5,830 skill packages. The method is execution-free, producing no output tokens and invoking no tools, with a compact implementation running in 259-284 ms median.

Why it matters: This paper presents a concrete, recent AI safety technique with strong empirical results and practical efficiency, valuable for developers and researchers working on agent security.

arxiv2d agoScore 85

AI safetyLLM agentsmalicious skill detection

Research Editor selected

PyRUA-DataFarm: LLM Agent as Cost-Efficient Teleoperator for Robot Data Collection

PyRUA-DataFarm uses an LLM agent to automatically rebuild robot tasks in simulation, generating real-ready training data without teleoperation. Policies trained on this farm data match or exceed those trained on teleoperated episodes across five real-robot tasks, with lower cost and better robustness to visual changes. The system includes automated certification of expert controllers and rendered episodes.

Why it matters: This presents a concrete, recent AI development with measurable cost and performance gains, offering practical value for robot learning and data collection.

arxiv2d agoScore 85

robot learningdata generationLLM agents

Research Editor selected

Think with Skills: Pseudocode Planning Before Execution for Advanced Reasoners

The paper introduces Think with Skills, a method that trains reasoning models to convert relevant skill descriptions into task-specific pseudocode plans before execution. The pseudocode explicitly represents operations, conditions, intermediate results, and dependencies, bridging declarative skill descriptions and agent actions. Training combines supervised fine-tuning on tasks, skills, pseudocode plans, and execution trajectories with Group Relative Policy Optimization. On SkillsBench, the method raises task success from 35.0% to 62.0%.

Why it matters: It presents a concrete training approach that substantially improves how language-model agents proceduralize and coordinate expert-written skills.

arxiv3d agoScore 82

LLM agentsplanningreasoning models

Research Editor selected

BLPO: Beyond-Local Policy Optimization for Long-Horizon LLM Agents

BLPO is a new reinforcement learning method for long-horizon LLM agents that estimates task-progress advantages from shared experience across tasks and updates, avoiding the need for a learned critic or extra rollouts. It summarizes trajectory prefixes into task-progress states and abstracts actions by function, using a global memory with balanced aggregation to control bias. Experiments on ALFWorld and WebShop with 1.5B and 7B policies show improved task success and faster learning under matched training budgets.

Why it matters: This paper introduces a concrete algorithmic improvement for training LLM agents on long-horizon tasks, with empirical gains on standard benchmarks, making it valuable for researchers and practitioners.

arxiv3d agoScore 85

LLM agentsreinforcement learninglong-horizon tasks

Research Editor selected

FedReady: Bounded LLM Agents Automate Federation Formation in Federated Learning

FedReady is a system that uses bounded LLM agents to automate federation formation in federated learning, including site discovery, eligibility verification, and preparation of shared contracts, local adapters, and task code. In simulations with 39 retinal sites and four tasks, it achieved perfect recall with 0.65 precision, removed all false positives, and supported 45,575 task-record instances across 100-round jobs. The system also demonstrated high fidelity to manual reference analysis with a maximum survival-curve difference of 0.001133.

Why it matters: This paper presents a concrete system that addresses a practical bottleneck in federated learning, with quantitative results showing feasibility and reliability, making it valuable for researchers and practitioners.

arxiv3d agoScore 82

Federated LearningLLM AgentsData Preparation

Research Editor selected

In-Model Context Compression with Agentic Soft Tokens for Long-Horizon LLM Agents

The paper introduces Agentic Soft-Token Compression (ASTC), a training-free method that uses soft tokens within an LLM agent to compress accumulated context for long-horizon tasks. ASTC leverages the model's native recurrent dynamics and hybrid Gated Delta Network (GDN)-attention to carry forward absorbed information as agentic soft tokens while retaining exact values as hard tokens. Experiments show ASTC matches or exceeds uncompressed accuracy with up to 50.68% fewer attention pairs, and trained compressors require 15.4 to 103.7 times more attention pairs per successful task.

Why it matters: This research presents a novel, training-free approach to context compression that significantly reduces computational cost while maintaining accuracy, offering practical value for deploying LLM agents in long-horizon tasks.

arxiv4d agoScore 85

LLM agentscontext compressionsoft tokens

Research Editor selected

KEEPER: Learning Control-Sufficient Textual Memory for Long-Horizon LLM Agents

The paper introduces KEEPER, a framework that analyzes auxiliary objectives for recursive textual memory in LLM agents, identifying self-consistency bias where incorrect claims persist in memory. It compares four supervision targets and finds reward-related prediction achieves the highest task performance before and after offline reinforcement learning. A claim-level analysis reveals error persistence despite corrective observations, motivating evaluation of error persistence and valid-information retention alongside task outcomes.

Why it matters: Provides a concrete, recent research contribution addressing a critical failure mode in long-horizon LLM agent memory, with practical implications for agent reliability.

arxiv4d agoScore 85

LLM agentsmemory managementreinforcement learning