The actual paper: https://arxiv.org/abs/2511.08425v3 https://arxiv.org/abs/2511.08425v3 > Our key insight is to leverage numerical optimal control to steer the

The actual paper: https://arxiv.org/abs/2511.08425v3 https://arxiv.org/abs/2511.08425v3 > Our key insight is to leverage numerical optimal control to steer the sampling trajectory so that constraints are satisfied precisely at the terminal time. Doesn't seem so "fool proof" to me as where the inevitable media spin will take it. Then, how do you know "where" to steer weights? "Safe" has no agreed upon definition