Verifiable Rewards Improve Native Visual Reasoning Training
A 300-task procedural suite replaces preference-only judging with task-specific scorers, raising matched reinforcement-learning performance from 0.509 to 0.548.
Uniform Loss Filters Valleys Through Velocity Contrast
A non-Hermitian acoustic crystal uses ordinary background dissipation to reach switchable valley polarization without tailored excitation or external fields.
