Testing and fine-tuning a LLM from the outside, without interacting with the backend. Concepts Prompt Injection In Context Learning Prompt Chaining Iterative Self-Evaluation