Trenton Bricken

Member of Technical Staff, Alignment Science at Anthropic

Bricken helped make sparse autoencoders a practical tool for finding interpretable features inside language models. He now develops automated audits that look for hidden objectives and other signs of model misalignment.