Nam Nguyen

Risk-sensitive planning with MCTS is now possible, with new algorithms achieving provable regret bounds and outperforming mean-optimal policies in avoiding catastrophic outcomes.

A single token can bypass most safety filters.

Conservation laws reveal how modern neural architectures maintain implicit biases during training, shedding light on their remarkable performance.

Exploiting Hamiltonian symmetries can significantly reduce the computational burden of electronic structure calculations, offering a more efficient path to accurate molecular simulations.