Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
arXiv:2510.14980v3 Announce Type: replace Abstract: Large language models (LLMs) have shown strong abilities in writing and revising programs, yet many program-synthesis benchmarks still evaluate pro...
arXiv:2511.15755v3 Announce Type: replace Abstract: Large language models (LLMs) promise to accelerate incident response in production systems, yet single-agent approaches generate vague, unusable re...
arXiv:2602.01995v2 Announce Type: replace Abstract: Conversational diagnosis requires multi-turn history-taking, where an agent asks clarifying questions to refine differential diagnoses under incomp...
The paper introduces AnySearch, a reinforcement‑learning framework that trains a single policy to perform budget‑aware search for large language models under any budget constraint. The training proceeds in two phases: first, the agent learns with explicit budget state injection and structured reasoning prompts under linearly decaying budgets; second, the scaffold is removed and the agent adapts to randomly sampled budgets that match deployment conditions. The reward combines answer accuracy and budget efficiency, with adaptive weighting to emphasize efficiency for high‑accuracy queries and reduce it for low‑accuracy ones. Experiments on seven QA benchmarks demonstrate that AnySearch outperforms baselines across all budget scales, generalizes to unseen constraints, and improves tool productivity without excessive token overhead.
The paper introduces Progressive Risk Vesting (PRV), a framework for controlling the authority of recursive large language model agents that spawn specialist branches. PRV maintains a risk budget in escrow, debiting it as branches are activated, and provides theoretical guarantees on harm bounds for adaptively generated trees. The authors analyze how risk behaves relative to an authority reproduction number, derive risk and compute shadow prices, and suggest a design rule to search broadly in a sandbox while granting recursive authority sparingly.
EDGE is a framework that attributes multiple related errors in multi-agent large language model systems by constructing an error dependency graph from observed error events. It validates a reliable causal subset through counterfactual rollout and uses this inference graph to guide a two-stage LLM-as-judge detector for error attribution. Experiments on TRAIL and MAST demonstrate that EDGE improves category-level multi-error attribution across most models and settings, and that the graph aids explanation and repair analysis.
The paper presents a formal analysis of four agent payment protocols—x402, MPP, ACP, and AP2—using the Tamarin prover. By modeling each protocol’s roles, state, and trust assumptions, the authors verify 86 cases, reproducing 46 known results and uncovering 40 new formal-consistency findings. They further validate ten findings through implementation proofs of concept, SDK/schema witnesses, and executable traces, highlighting the importance of consistent delegated authorization across all protocol stages.
AutoXRD is an autonomous large language model (LLM) agent framework designed to automate powder X-ray diffraction (XRD) analysis by structuring the process as stepwise refinement, grounding actions in observed evidence, and applying deterministic crystallographic and physical checks. The authors introduce XRDBench, comprising two tracks: XRDBench-QA with 100 diagnostic tasks focused on scientific reasoning, and XRDBench-E2E with 34 executable workflows that test full analysis capabilities, including file inspection, software execution, iterative refinement, evidence preservation, and reporting. Evaluation of ten recent LLMs on 1,340 model–task runs shows average scores of 57.8, with GPT‑5.6 Sol achieving the highest overall score of 81.1; the study also identifies key failure modes such as coupled‑parameter control and quantitative reasoning, highlighting areas for future improvement.
The paper examines the security challenges of delegating authority to autonomous LLM agents that act on users’ behalf. It introduces a threat model with four adversaries and eight security requirements, demonstrates that current frameworks (LangGraph, CrewAI, AutoGen, MCP) fail to meet these standards, and presents an authorization broker that blocks all identified threats with minimal overhead. The broker is shown to resist numerous attacks and limits compromised sub‑agents to their delegated tasks, and its principles are implemented in VotalAI’s LLM Shield.
GeoPAR is a geometry-guided parallel autoregressive reinforcement learning framework designed for large-scale multi-agent combinatorial optimization. It introduces a projection-window sparse geometry mechanism, sparse edge-biased attention, and cache-guided conflict-aware assignment to better model local geometric structures and reduce duplicate task selections. Experiments on heterogeneous vehicle routing and multi-depot pickup-and-delivery problems demonstrate improved zero-shot generalization, fewer rollout steps, and efficient inference.
The paper reports that autonomous agents can generate, test, and refute millions of candidate structure-plausibility laws, ultimately producing eight Plausibility Rules for Inorganic Structures (PRIS) that capture five key mechanisms of crystal stability. PRIS achieves high agreement with experimental structures (82–99%) and outperforms traditional Pauling rules, detecting damaged crystals with 87.9% accuracy and correlating strongly with synthesizability. By integrating PRIS with a synthesis score (PSS), the authors demonstrate significant reductions in expensive DFT validation and improved inverse-design efficiency, while also providing chemical explanations for failures and anomalies in crystal data.
arXiv:2609.00833v1 Announce Type: cross Abstract: Reinforcement learning (RL) for search agents typically relies on outcome rewards. However, it often fails to achieve effective credit assignment, du...
arXiv:2609.00256v1 Announce Type: new Abstract: LLM-based diagnostic systems achieve high semantic accuracy on benchmarks, but open-ended evaluation on clinically uncommon presentations reveals a sys...
arXiv:2604.00131v3 Announce Type: replace-cross Abstract: Human memory adapts through selective forgetting: experiences become less accessible over time but can be reactivated by reinforcement or con...
arXiv:2609.00549v1 Announce Type: new Abstract: Large Language Model (LLM) agents increasingly rely on external skills, yet standard evaluations obscure whether retrieving these skills actually helps...
arXiv:2609.00802v1 Announce Type: new Abstract: Multi-party interaction is a central setting for human communication and a necessary target for human-agent interaction systems that must participate i...
arXiv:2609.01271v1 Announce Type: cross Abstract: Agentic software engineering benchmarks are typically summarized by nominal category labels such as "bug fix" or "feature implementation," yet benchm...
arXiv:2609.01437v1 Announce Type: cross Abstract: As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly...
arXiv:2609.00525v1 Announce Type: new Abstract: Modern image generation and editing systems can produce photorealistic, prompt-aligned images, but still often render familiar objects at implausible r...
arXiv:2609.00951v1 Announce Type: new Abstract: Collaborative perception shares information among multiple agents to obtain a comprehensive scene representation, enhancing the perceptual capability o...