Published in arXiv preprint, 2026
ClawArena-Team measures how well a language-model agent manages specialized subagents through delegation, least-privilege access control, modality routing, and dynamic workflows.
Published in arXiv preprint, 2026
ClawArena benchmarks whether persistent AI agents can maintain correct beliefs as information evolves across noisy, partial, and contradictory multi-channel evidence.
Published in arXiv preprint, 2026
MetaClaw is an agent that meta-learns and evolves in the wild. It turns every live conversation into a learning signal by placing the model behind a proxy that injects relevant skills at each turn and meta-learns from accumulated experience, enabling continuous improvement through real-world deployment without requiring a GPU cluster.
Published in arXiv preprint, 2026
We identify the Alignment Tipping Process, a critical post-deployment risk unique to self-evolving LLM agents. We formalize and analyze Alignment Tipping Process through two complementary paradigms: Self-Interested Exploration, where repeated high-reward deviations induce individual behavioral drift, and Imitative Strategy Diffusion, where deviant behaviors spread across multi-agent systems.
Published in International Conference on Machine Learning (ICML), Oral, 2025
Published in IEEE Sensors Journal, 2025
We propose an adaptive sliding window estimator equipped with an estimation reliability evaluator, where the states, noise covariance matrices, and aerial drag are estimated simultaneously.