LLMs on Drugs: Language Models Are Few-Shot Consumers

Alexander Doudkin

Large language models (LLMs) are sensitive to the personas imposed on them at inference time, yet prompt-level "drug" interventions have never been benchmarked rigorously. We prese…