Who's Harry Potter? An interactive walk through approximate unlearning in LLMs

An interactive marimo notebook explaining Eldan & Russinovich's 2023 paper "Who's Harry Potter? Approximate Unlearning in LLMs." The headline visualization frames the unlearning mechanism as edge surgery on the model's knowledge graph: the algorithm removes specific learned associations between entities while leaving the surrounding language graph intact. Runs entirely in the browser via Pyodide.