Language models generate maximally activating examples for any residual direction

Researchers develop Maximally-Activating Example Metamodels (MAEMs), language models that map a...