Carly Taylor compares creating human communication undecipherable by AI to attempts to poison LLM context or trick facial recognition systems