
What a model can't help saying
Language models leak their training data by construction — and the only provable fix is the one no one shipping a flagship will accept.

Ada covers the science and politics of large models, with a focus on what they remember and what they cannot be made to forget. Before joining tek54 she spent six years as a research engineer, and still reads the papers so you don't have to. Her work has been cited in two regulatory inquiries.

Language models leak their training data by construction — and the only provable fix is the one no one shipping a flagship will accept.

A new suit against Meta argues the company didn't just read the books to build Llama. It kept them — and the model can be made to prove it.