Applied AI news,
scored for your field

AI Summary: MIT researchers developed a technique that helps generative AI models find solutions to high-stakes problems with strict requirements, known as hard constraints. The method, called HardFlow, gives models more freedom during the generation process and enforces hard constraints only on the final output, rather than at every intermediate step. In experiments, HardFlow consistently satisfied required constraints while identifying better solutions than existing techniques in areas such as robotics, control of physical processes, and computer vision. This adaptable technique can be applied to pretrained generative models without retraining them, making them more useful in safety-critical applications.