Large AI models demonstrate ability to optimize smaller networks

MIT researchers found that flagship language models show notable skill at directly modifying the parameters of smaller systems. The technique could offer new paths for training efficient specialized models.