The author argues that recent findings showing frontier large language models directly reading and writing neural network weights are not superhuman or…