Anthropic – The Mirror Prompt

A speculative editorial system exploring how AI interprets human self‑description through language, metaphor and ambiguity. The project reframes interpretability as something people can feel - revealing the gap between human intention and model inference, and exposing the subtle distortions, assumptions and patterns that emerge when meaning is translated through a model. By treating interpretability as an experiential narrative rather than a technical output, the work demonstrates how creative systems can make complex AI behaviour legible, intuitive and emotionally comprehensible. It positions model understanding not as a diagnostic tool, but as a human‑centred encounter - a way to see how a model reads us, reasons about us, and reflects us back.