SeePhys Pro: Benchmarking Multimodal RLVR in Physics Reasoning
Explore SeePhys Pro, a new benchmark analyzing modality transfer and blind training effects in multimodal reinforcement learning for physics reasoning.
Value of Brain Data in Machine Learning Models
Explore how brain data enhances machine learning models, its value vs task data, and optimal collection strategies in NeuroAI research.
How AI Learns Preferences from Learning Agents
Explore how AI infers human preferences from learning agents using inverse reinforcement learning and no-regret models for better alignment.
Temporal Knowledge Drift in LLMs: Geometry of Forgetting
Explore how temporal knowledge drift impacts LLMs and why it's a fundamental geometric issue, not just an engineering flaw.
Online Trajectory Verification Boosts AI Skill Distillation
Discover how online trajectory verification with PDI and SPARK improves AI skill distillation, boosting performance and cutting inference costs.