Factual knowledge in large language models is stored within the network connections between neurons rather than inside isolated individual neurons