Using and trusting LLMs: evaluation, RAG, agents, and safety.
Benchmarks, contamination, judges, and pass@k.
Look the answer up, then write it.
A model that calls functions and comes back.
Jailbreaks, prompt injection, and poisoning — the ideas.
What the model stored, and when that is a problem.
Inspired by the work of Mohit Iyyer, Umar Jamil.