PhD Fellowship, Mechanistic Interpretability for Large Language Model Security

Copenhagen, Denmark

• In this fellowship, you'll research mechanistic interpretability methods to improve LLM security and mitigate false information attacks.

• Develop novel mechanistic interpretability methods and evaluation protocols across different model lifecycle stages.

• Collaborate with supervisors, postdoctoral researchers, and external partners to advance mechanistic interpretability.

• Author and publish research papers in high-impact venues and disseminate findings nationally and internationally.

• Undertake PhD courses in academic writing and specialised ML/NLP topics, and conduct research abroad.