Kids outlearn AI—and we still don’t know why

Children achieve fluency after encountering far less language than large AI models require. Researchers are using small language models, child-recorded video, and interactive experiments to investigate this data efficiency gap and improve AI for languages with limited training material.

儿童接触的语言远少于大型人工智能模型所需的训练数据，却能熟练掌握语言。研究人员正利用小型语言模型、儿童视角录像和互动实验研究这种数据效率差距，并希望改善训练材料有限的语言所使用的人工智能。