Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
We evaluate whether AI agents can match the reported findings of published biomedical studies using public data. Existing evaluations do not consistently separate analysis from prior knowledge or retr...
LLM orchestration investigates how an orchestrator coordinates a group of autonomous agents to achieve common goals or maximize collective welfare. The agents are typically heterogeneous, each holding...
In clinical practice, agreement among independent experts is treated as evidence of reliability, and multi-round consensus has become a core mechanism of agentic medical question-answering systems. Wh...
The article explains the transition from the original Cowork system, which ran model inference and tool calls in a local Anthropic VM, to a new cloud‑based version that hosts both inference and the VM. The new design isolates each session in its own sandbox, eliminates local resource drain, and allows file access through the desktop app, addressing concerns about battery usage, performance, and continuity when a laptop is closed.
Agent tasks require sequences of interdependent decisions. Autoregressive models support more flexible decision interfaces than conventional classifiers but incur the latency of token-by-token generat...
T-Search is an open-weight agentic retriever designed for hard multi-step search tasks. It performs bounded multi-round searches over a fixed corpus, returning ranked evidence chunks with brief justifications, while allowing the answer generation component to be swapped without retraining. Trained on synthetic search tasks and evaluated on seven English and Russian benchmarks, it achieves 56.0 Recall@10 with one rollout and 61.3 with three fused rollouts, surpassing larger open models.
The paper investigates how frontier AI systems perform on tasks related to cyber security, specifically exploit generation, vulnerability repair, and subsequent attacks, using five nonpublic software environments. It evaluates both open‑weight and proprietary models, employing deterministic graders rather than LLM judges to score performance. Results show significant variation across systems and vulnerability types, with repair scores higher than attack scores in two environments and lower in three, and highlight that passing an initial security test does not guarantee long‑term defense, as 92 of 524 defender test intervals still experienced successful exploits after the first one was stopped.
The paper proposes a new method for generating synthetic survey responses that uses declared aggregate preference anchors to create group-indexed choice policies, avoiding the mixing of inference-time information with pretraining associations. It introduces a parameter‑efficient adapter trained via Direct Preference Optimization to map GPS coordinates to synthetic responses, and evaluates these adapters on World Values Survey items, showing they recover imposed labels and exhibit strong correlation with GPS trust scores. The study demonstrates that anchored policies can preserve aggregate signals while not necessarily mirroring human response patterns, offering an inspectable construction and evaluation framework that separates anchor transfer from human criterion agreement.
The paper introduces SciKGExtract, a schema-guided framework that uses large‑language‑model extraction, chemical normalization, and agent‑based evaluation to convert heterogeneous materials science literature into a knowledge graph. Applied to 176 atomic‑layer‑deposition papers on ZnO and IGZO, the system improves extraction F1 from 0.591 to 0.805 for ZnO with agentic refinement, while IGZO remains more challenging at 0.344. Evaluation against a detailed schema of 65 experimental properties and 155 quantitative nodes reveals segmentation and numerical assignment errors, highlighting the complementary role of chemical canonicalization and agentic verification in producing machine‑actionable experimental knowledge.
The paper introduces Guidance‑TTT, a method that separates strategic planning from execution in test‑time training for large language models. A small guidance model is trained at test time to propose high‑level changes, while a frozen, larger execution model implements these changes, reducing the cost of maintaining gradients and optimizer states. Guidance‑TTT achieves strong results across four domains—combinatorial optimization, heuristic programming, machine learning, and GPU kernel optimization—outperforming prior work and matching state‑of‑the‑art leaderboard scores.
MoCAR (Motion-code Coordinate-aware AutoRegression) is a decoder‑only framework that treats continuous trajectory forecasting as next‑code prediction in a coordinate‑aware latent space. It learns a continuous motion‑code space from endpoint‑normalized trajectory segments, using historical codes as a teacher‑forced prefix and generating future codes autoregressively while updating local scene context. The method achieves top‑tier performance on Argoverse benchmarks, transfers well from AV2 to AV1 in zero‑shot evaluation, and improves on turn‑heavy scenarios without requiring trajectory‑space re‑tokenization or proposal‑and‑refinement pipelines.
The paper introduces a reinforcement learning framework for optimizing workload‑aware power delivery networks (PDNs) in VLSI chips. By mapping system‑level power traces to spatial power density distributions, the method adaptively allocates PDN resources and uses a Deep Q‑Network to perform wire‑width optimization, reducing PDN area while meeting electromigration and IR‑drop constraints. Experiments on multiprocessor floorplans with PARSEC and SPLASH‑2 workloads show a 47% area reduction and a 26× speedup over simulated annealing.
The paper introduces TrustMI, a method to causally control how large language model assistants decide to trust their users. By creating 2,000 contrastive conversations that vary in ability, benevolence, and integrity, the authors learn steering matrices that adjust trust decisions along linear directions in model activations while keeping the model parameters frozen. Experiments across six instruction‑tuned models show that these steering changes reliably alter trust decisions and affect safety‑related behaviors such as compliance with harmful requests, prompt injections, and insider threats.
RocketAgent is a long‑horizon engineering agent designed to streamline the multidisciplinary preliminary design of liquid‑rocket thrust chambers. It employs a single plan‑owning Coding Agent to coordinate tasks such as performance sizing, subsystem optimization, geometry generation, and multiphysics assessment, while a provenance‑aware knowledge graph and a typed Design Intermediate Representation manage method selection and shared design data. Revision‑aware checks invalidate affected results and block superseded inputs, ensuring that changes are properly approved and that subsequent design steps are coordinated, as demonstrated in a representative simulation‑based design and controlled scenario tests.
The paper introduces ARRO, a reinforcement‑learning framework that enforces minimal‑change editing by auditing source images, instructions, and edited outputs for unimplemented and unintended changes. Using a vision‑language reward model and a group‑level rubric, ARRO achieves higher EditScore on multiple benchmarks and reduces off‑target pixel changes by 8.4% on FLUX.1 Kontext‑dev. The approach eliminates the need for per‑instruction human annotations and demonstrates strong performance across several evaluation sets.
ThunderSyncRL is a training framework that eliminates idle time in agentic reinforcement learning by starting gradient computation immediately once all necessary inputs are available, thereby avoiding policy staleness. It applies to group relative policy optimization (GRPO) by computing trajectory score gradients as soon as rewards arrive, and to on‑policy distillation (OPD) by updating gradients for completed agentic turns while tool calls execute. Experiments on SWE‑bench Verified and Terminal Bench 4.0 show that ThunderSyncRL matches synchronous training performance up to 1.9× faster and outperforms asynchronous training by up to 2.47 percentage points at a fixed budget.
The paper investigates the challenges of optimizing compacted context models, particularly the KV cache, in continual learning scenarios. It identifies the optimization landscape as brittle and flat, and proposes a simplified Perceiver-based architecture that matches or surpasses full Perceiver transformers in continuous context compaction. Experiments on MCQ tasks in Finance, Legal, Gutenberg, and Code demonstrate the effectiveness of this approach.
UniIntervene++ is an adaptive intervention agent that learns to balance autonomous execution and assisted behaviors during online reinforcement learning. It models the evolving policy, trajectory correction, and task-structured CodePolicy as Options in a unified semi‑Markov decision process, learning their values online. The agent periodically probes the policy to keep control allocation responsive, and uses coupled experience learning to improve both the policy and future interventions, achieving an 89.67% success rate on five real‑world tasks while cutting human intervention by 94.6%.
The study explores how the language used for reasoning affects retrieval‑augmented generation (RAG) in a monolingual German setting. Using a German RAG question‑answering testbed based on the tabletop game The Dark Eye, the authors show that aligning the reasoning language with the query and retrieved documents improves performance, with German reasoning outperforming French reasoning. However, German reasoning still does not surpass the model’s native English reasoning, indicating that native multilingual reasoning is necessary for optimal results.
AdaStep introduces an adaptive step-credit weighting technique for agentic reinforcement learning, addressing the coarse granularity of trajectory-level objectives in long-horizon LLM agents. By formulating the weighting as a mean-squared-error estimation problem and deriving an optimal per-state shrinkage coefficient, AdaStep selectively preserves local credit when return variation is due to the chosen action and suppresses it when downstream randomness dominates. The method requires only lightweight scalar computations, no critic or extra rollouts, and demonstrates consistent performance gains across three model backbones on ALFWorld, WebShop, and ScienceWorld.