Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
arXiv:2606.11350v2 Announce Type: replace Abstract: Retrieval-augmented generation (RAG) degrades when applied to large, heterogeneous document collections. In RAG systems, dense similarity becomes l...
arXiv:2606.17029v2 Announce Type: replace Abstract: Deep research agents synthesize long-form reports by searching and reasoning over retrieved evidence. Reinforcement learning with rubric-based rewa...
arXiv:2505.13546v2 Announce Type: replace-cross Abstract: Automatic prompt generation plays a crucial role in enabling general-purpose multi-agent systems to perform diverse tasks autonomously. Exist...
arXiv:2606.16710v2 Announce Type: replace-cross Abstract: Multi-agent systems (MAS) are credited with solving complex problems, yet their agents routinely act on context that may be false. Prior work...
arXiv:2608.12313v3 Announce Type: replace-cross Abstract: Agentic video creation systems remain far from cinematic-grade quality: their underlying models struggle to plan how story, characters, visua...
arXiv:2610.20240v1 Announce Type: new Abstract: A multiplayer world model must predict how the independent actions of several players change a shared environment, including changes outside a player's...
arXiv:2610.20384v1 Announce Type: new Abstract: Agents for mixed-degradation image restoration typically learn what a tool does to an image only after running it: failed calls waste computation and h...
arXiv:2610.20571v1 Announce Type: new Abstract: Long-form agentic video generation must turn a screenplay into shots that remain spatially and temporally consistent across nonconsecutive requests. Pa...
arXiv:2610.20584v1 Announce Type: new Abstract: Computer-use agents have demonstrated strong capabilities in interpreting screens and executing keyboard and mouse actions. However, games, with their...
arXiv:2610. 19179v1 Announce Type: cross Abstract: We study packet routing in queuing systems where queues act as selfish, myopic agents that use no-regret learning algorithms to select servers with unknown service rates.
arXiv:2610.18038v2 Announce Type: replace-cross Abstract: World action models (WAMs) built on pretrained video generation models adapt quickly to new robotic tasks, but they forget previously learned...
L3M is a runtime built in Lean 4 that enables self‑extending and self‑improving agents to operate across humans, models, and code without requiring human attention in every loop. It stores all computation state in Git, allowing rich state trajectories that can be resumed, rewound, forked, compared, and merged. The system uses reversibility and kernel‑checked theorems to ensure safety, providing agents with sealed Lean programs as tools that can be reviewed and undone when necessary.
StateCraft introduces a new architecture that separates a persistent, updatable abstract state from the visual state in video world models. By pairing a language model with a video diffusion model, it tracks entity changes over time and conditions visual generation on this abstract state. The approach is evaluated on generative interfaces and a new benchmark, showing improved generation quality, memory usage, and controllability, while enabling direct editing of unseen states and extending to spatial 3D environments.
The paper investigates how large language model (LLM) workflows and agents make binary decisions—such as allowing or blocking an input—when human deferral is not possible. It proposes using System One decision scores that default to a conservative fallback in uncertain cases and control false omission risk within the subset of inputs that receive the utility‑maximizing action. Experiments on six datasets spanning moderation, safety, and reranking show that naive threshold baselines often exceed target risk, whereas a Learn‑Then‑Test approach achieves only 7.8% risk, demonstrating the limitations of probability‑like interfaces and the benefits of risk‑budgeted selection.
The paper explores Behavior Cloning (BC) for autonomous driving using bird's‑eye view (BEV) inputs in simulated urban settings. It introduces a weighted BC approach that employs a kernel density estimator to adjust training sample weights, emphasizing rare but critical actions like stopping at red lights and accelerating at green lights. Experiments in the CARLA simulator show that this weighted method improves route completion and generalizes better to unseen test environments compared to standard BC.
The paper introduces MMLA‑RSI, a recursive self‑improvement framework within a Memory‑Mediated Learning Architecture. It presents two concrete implementations: Reasoning‑Time Training (RTT), which updates the policy state, and Reasoning‑Time Memory Adaptation (RTMA), which modifies the authoritative memory. Both routes treat their own adaptation mechanisms as improvable objects, enabling successive generations of agents to refine policy and memory through bounded rule languages and shared interfaces.
The paper introduces HiRe-MoMa, a hierarchical residual policy learning framework that fine‑tunes a frozen base policy for mobile manipulation. It uses a policy‑agnostic residual split into mobile‑base and manipulator‑arm heads to reduce gradient interference, and incorporates expert‑guided value alignment to improve sample efficiency with sparse human interventions. Experiments on the TIAGo++ platform show that HiRe‑MoMa boosts task performance after only about 30 minutes of online interaction, regardless of the underlying base policy.
The paper introduces Byrd-NAFL, a Byzantine‑robust federated learning algorithm that incorporates Nesterov momentum at the server while keeping workers stateless. It applies any soft γ‑Byzantine resilient aggregation rule and provides finite‑time convergence guarantees for smooth non‑convex losses across all momentum factors β∈[0,1). Experiments on MNIST, CIFAR‑10, and Sentiment140 demonstrate faster convergence and higher accuracy than existing benchmarks under various attacks.
LatentRemedy is a method designed to improve inference efficiency in multi‑agent systems that share non‑prefix KV caches. By inserting a small number of task‑conditioned latent tokens generated via an MLP, the target agent can better interpret reused context without recomputing the shared block. Experiments demonstrate that LatentRemedy consistently outperforms existing reuse baselines on conflicting tasks while keeping TTFT close to direct reuse as cache size grows.
The paper proposes a cost‑efficient privacy enforcement method for email‑generating large language model (LLM) agents that intercepts drafts, creates an information‑release contract, rewrites the draft locally with a Gemma rewriter, and then optionally calls a final PrivacyChecker. Experiments on normal and strategically malicious prompts show that the Gemma + Checker approach reduces leakage from 11.41 % to 4.03 % and improves usable output rates, while also cutting the cost amplification of checker‑only pipelines from 1.32× to 1.02×. The results demonstrate that moving repair to a cheaper local layer can significantly lower both privacy leakage and operational cost, yet still rely on a checker for high‑risk sends.