Fei-Fei Li: AI safety evaluation should not be the sole responsibility of developers; independent evaluation standards needed

PANews, September 23 - According to Bloomberg, Fei-Fei Li, co-founder of World Labs and professor at Stanford University, said that increasingly capable artificial intelligence systems should not have their safety evaluated solely by the companies developing them. Companies can internally test a model's capabilities and safety properties, but academia, government, and industry still need to jointly establish broader independent evaluation standards. She believes that the potential risks of AI depend on how humans develop, govern, and use the technology, and that humans must retain control. Li also said that World Labs is developing "world models" that can generate and reconstruct three-dimensional environments, and the related technology can be used in scenarios such as robot training and evaluation.