Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
An agent that uses tools typically responds to what the user explicitly asks, yet completing the task may require information the user never requested. Work on proactive agents mainly studies whether...
Modern coding agents can deliver increasingly large repository-level changes, and recent benchmarks reflect this by emphasizing long-horizon tasks with large reference implementations. Many benchmarks...
Skill evolution improves the capabilities of large language models by analyzing trajectories generated under a given skill and modifying the skill accordingly. Existing approaches typically generate a...
LLM-based multi-agent systems (MAS) increasingly use latent collaboration to avoid the information loss and repeated encoding-decoding overhead of natural-language communication. However, directly for...
High-resolution visual question answering often fails because a multimodal model does not acquire the small, spatially localized evidence needed to answer a question. Sequential zooming can recover de...
Multi-agent LLM systems are increasingly structured like organizations, with roles, critics and shared memory, yet they are evaluated on tasks that last minutes. We ask how such an organization behave...
Recursive self-improvement (RSI) aims to achieve compounding gains by having models improve themselves. While most existing RSI systems optimize external agent harnesses or prompts around a frozen bas...
The paper introduces robust successor features, a method that extends the successor representation to handle uncertainty in both reward functions and transition kernels within linear Markov Decision Processes. It provides a theoretical bound on Generalized Policy Improvement that quantifies performance loss due to mismatched dynamics, and demonstrates the approach on grid-based benchmarks against prior methods that consider only reward or transition differences.
The paper introduces Gradient‑Momentum Coupling (GMC), a method that quantifies learning progress by measuring how strongly a sample influences changes in the parameter space, using the normalized absolute product of its gradient and the momentum of previous gradients. GMC filters out noise by accumulating consistent directions of change while canceling random fluctuations, leading to a more uniform prioritization across tasks with varying noise levels and better ranking of learnable tasks by improvement speed. Experiments on MiniGrid MultiRoom tasks show that replacing prediction error with GMC in the Intrinsic Curiosity Module restores exploration capabilities that were lost to unpredictable observations.
The paper introduces TRACE, a fine‑tuning framework for Retrieval‑Augmented Generation (RAG) that addresses conflicts between retrieved knowledge and a model’s internal knowledge. TRACE uses multi‑agent debate traces to identify correct and incorrect candidates and answer‑shift patterns, providing fine‑grained supervision for reliable knowledge‑source selection. It also incorporates an answer‑completeness regularization mechanism to prevent empty, overly short, or prematurely terminated responses, thereby improving robustness against misleading retrieved content and enhancing answer quality.
The paper introduces TalkMesh, a decentralized network of small language model agents that learn to communicate effectively during inference. Each agent proposes an answer, scores it with a confidence head, and the most confident agent broadcasts a hint; lower‑confidence agents revise their proposals if a new suggestion scores higher. This gossip‑based consensus, trained via group relative policy optimization, enables a mesh of three agents to match the accuracy of majority voting over 32 samples, and scales to larger meshes to significantly boost performance on benchmarks like GSM8K and MATH-500.
The paper introduces Morphogene, a compact latent blueprint that links an agent’s body and control policy, enabling coordinated changes at the limb level through AdaConcat. Building on this, GeCode treats co-design as exploration within Morphogene space, using local refinement and global exploration to efficiently navigate design regions. Experiments on various 2D and 3D tasks show that GeCode outperforms state‑of‑the‑art methods, achieving faster convergence and higher performance.
HySTAR is a MAPPO-based framework that addresses structural target drift in cooperative multi‑agent reinforcement learning by anchoring an overlapping sparse hypergraph as a stable high‑order value‑decomposition scaffold. It separates adaptive representation learning from a temporally consistent decomposition basis, using a spatiotemporal encoder to capture physical and task‑dependent interactions and combining temporal and structural relevance to compute agent‑specific advantages. Experiments on SMAC, GRF, Traffic Junction, and MPE show consistent improvements over MAPPO‑style, value‑factorization, and dynamic‑grouping baselines, achieving significant gains in performance and convergence speed.
The paper reports on applying AutoResearch—a large language model that iteratively edits training scripts—to optimize embedding systems for a book recommendation pipeline at production scale. Over twelve weeks, the authors ran 220+ experiments across two representation‑learning systems, uncovering five recurring failure modes (infrastructure fragility, agent memory decay, search‑direction stagnation, iteration‑cost asymmetry, and metric fixation) that were not present in smaller settings. They propose a three‑principle scaffolding (prevent, persist, redirect) to address these failures, achieving a 1.82× lift in Recall@6, a 2.1× lift in coherence, and an autonomous text‑only fallback that expanded catalog coverage by 5.8×.
A commercial incentive need not enter the final ranking algorithm to affect a shopping assistant's recommendation: it may instead influence which preference question the assistant asks. We make this d...
Simon Willison reflects on the rapid and unexpected advancements in AI capabilities, particularly in areas like cyber, swarming, and message boards. He emphasizes that security posture requires more than system hardening; it must be embedded in company culture and involve people adapting alongside technological changes. Willison urges organizations worldwide to assess their resilience to sudden AI jumps, ensuring people, systems, processes, incident response, and communication are prepared for such surprises.
The paper explores Retrospection-Only Fine-Tuning (ROFT), a method where a language-model agent improves its behavior by generating and training on explanations of its own experiences, without external teachers or reward signals. In software‑engineering tasks with Qwen3.5‑4B, ROFT achieves comparable or better solve rates than GRPO while requiring fewer updates and training time, and can learn from failures alone. Behavioral analysis shows ROFT indirectly assigns credit to actions and can produce shorter, more direct solutions when prompted to focus on direct solutions.
SEABench is a benchmark designed to study endogenous misalignment in self‑evolving large language model agents. It contains 48 longitudinal task sequences across various evolution surfaces, task domains, and harm types, and includes an adaptive trajectory discovery pipeline that probes for failures while preserving task intent. Evaluations show that self‑evolution improves task completion rates but often introduces safety failures absent in non‑evolving baselines, with divergent safety behaviors reflected in agents’ chain‑of‑thought reasoning that can be monitored to mitigate unsafe actions.
AutoBCI is an agentic framework that uses a Designer Agent and a Forecaster Agent to discover and select EEG decoding architectures across diverse tasks such as emotion recognition, motor imagery, and sleep staging. The Designer Agent performs Pool‑Guided Architecture Discovery (PGAD) to generate and refine models, while the Forecaster Agent uses Performance Estimation from Early Knowledge (PEEK) to predict full‑budget validation performance from early learning curves. In experiments on 14 EEG datasets, AutoBCI with Claude Opus 5.5 achieved a 64.16% average test balanced accuracy, slightly surpassing the best baseline, and PEEK reduced prediction error by 38.1% compared to the best-observed-score baseline.
The paper investigates how hybrid attention mechanisms—combining full softmax attention with recurrent alternatives—affect multilingual language models, especially for long sequences and poorly tokenized languages. Interpretability analysis reveals that cross‑lingual representations form patterns linked to the ordering of recurrent and full‑attention layers, with a notable spike in alignment after the first full‑attention layer. Distillation experiments show that alternative layer orderings consistently outperform the standard arrangement, achieving up to 2.5× faster learning, suggesting that starting with a full‑attention layer may benefit multilingual models.