Haibin Ling

Model selection for scientific forecasting is fundamentally protocol-dependent, with optimal choices reversing based on dataset size, system complexity, and evaluation metrics like state refresh versus closed-loop performance.

Physics-constrained generation now achieves comparable fidelity to multi-step methods in a single neural transport evaluation.

Instruction leakage can lead to misleadingly high accuracy in spatial relation tasks, revealing a critical flaw in goal-conditioned models that could misguide future research.