The idea of enforcing constraints only on the final output instead of every intermediate step is pretty interesting. I wonder how well this would translate to l

The idea of enforcing constraints only on the final output instead of every intermediate step is pretty interesting. I wonder how well this would translate to language models, where “safe” is much harder to define mathematically than a robot avoiding an obstacle.