NEC Laboratories America researchers introduce RILKE, a method that updates what a large language model knows without retraining. After 1,000 sequential edits, it kept near-perfect accuracy on edited facts and preserved general reasoning, while competing methods collapsed.