Design the user-facing reliability layer of any AI product: uncertainty and confidence display, provenance/citations, hallucination and sycophancy mitigations, explainability calibrated to stakes, and a named cause + recovery path for EVERY failure — refusal, timeout, content filter, context overflo
Design the user-facing reliability layer of any AI product: uncertainty and confidence display, provenance/citations, hallucination and sycophancy mitigations, explainability calibrated to stakes, and a named cause + recovery path for EVERY failure — refusal, timeout, content filter, context overflow, mic/STT death, TTS death, dead live-embed. Use when asked 'what happens when the AI is wrong/fail