Jannik Brinkmann reverse-engineers how language models represent concepts internally, and uses that understanding to make frontier AI systems more controllable and robust.