LLM Context Biases

Primacy bias, recency bias, sycophancy, and anchoring are predictable distortions in how LLMs weight information. Understanding them changes how you prompt, evaluate, and trust model outputs.