The mission? Teaching models to better understand how to build code will lead to superintelligent AI.

Scientists at Massachusetts Institute of Technology have devised a way for large language models to keep learning on the fly—a step toward building AI that continually improves itself.