This paper presents a predictive law for calculating the uplift of Large Language Model (LLM) ensemble performance based on diversity of thought. It provides an experimentally verified formal law and a compact heuristic for calculating uplift, which is tested on various datasets.

MoLE is a new framework for latent visual reasoning in vision-language models, which encourages different latent tokens to extract complementary visual information. It outperforms existing methods on five visual reasoning benchmarks, achieving an average score of 78.6.

A study of decentralized agent economies identifies recurring failures in the verification and settlement stages, where conforming work can remain unaccepted or valid evidence can be ignored. The research introduces guarantee closure, a criterion for determining whether guarantees established at one stage remain available and constrain later decisions.

The paper proposes a claim-anchored execution contract to address the issue of unverified claims in tool-agent auditing. The contract binds claims, evidence, and execution to ensure the integrity of tool usage. It exposes seven testable properties and achieves high attack-detection rates.

A study on agent evaluation reliability explores the impact of scaffolds and tasks on model rankings. The authors develop a Bayesian variance-decomposition framework to separate signal from noise in sparse, imbalanced leaderboards. They find that reliability depends on the measurement goal, scaffold choice can change conclusions, and more tasks may not resolve all uncertainty.

CompoWorld is a compositional environment scaling approach for training general agents. It composes a library of reusable services, enabling agents to connect information and actions across multiple services. This approach improves performance on benchmarks, surpassing frontier models.

Researchers propose AREX-2, a self-improving LLM agent that iteratively refines solutions through reflection and long-horizon execution. They demonstrate the agent's effectiveness on various benchmarks, showing potential for sustained improvement.

ContextRender is a context management method for LLM agents that uses a persistent graph of execution dependencies to track how earlier tool results are used in subsequent execution. It reduces inference cost by 10.2-32.2% while achieving task performance close to or above passing the full history.

FedLAFP proposes a framework for federated fine-tuning of pre-trained models, improving personalized prediction by using a combined low-rank and full-rank approach for adaptation and personalization.