ABot-C0 shows how data generation, generalist motion tracking, specialized locomotion policies, and runtime orchestration can turn quadruped robotics from a collection of demos into a more reusable product stack.
A new analysis explains how rollout staleness and learning rate jointly determine whether asynchronous RLHF remains stable, collapses locally, or drifts toward failure.
GRADE shows how a multi-agent AI system can improve reasoning by selectively activating experts, limiting communication, pruning weak branches, and recalibrating models when the roster changes.
A self-generated analogy game separates models that produce strong answers from those that can reliably detect errors—and offers a blueprint for evaluator governance without a standing answer key.
Agri-SAGE shows how retrieval, crop simulation, and different agent reasoning strategies create distinct trade-offs between simulated yield, adaptation, and computational cost.
MACR shows why reliable AI needs an explicit process for adjudicating conflicts, not another instruction to trust the model or the retrieved document.
A study of graduate bioscience trainees finds that reported LLM usage separates pre-course attitudes more consistently than prior AI coursework, offering a practical but limited signal for adaptive training.
A lifecycle framework reveals how AI systems can collect diverse views while quietly fixing the categories, proxies, and decision rights that matter most.
A bilingual feedback pipeline shows how organizations can turn messy comments into governed early-warning signals—without pretending the system has diagnosed bias.
MotionHalluc shows that multimodal models can sound like competent coaches while failing to verify which person actually performed which motion.