EPFL researchers have discovered key "units" in large AI models that seem to be important for language, mirroring the brain's language system. When these specific units were turned off, the models got ...