Exploring Model Medicine ideas for a broader audience.
We tried to diagnose Mistral 7B with Shell-Core Conflict Syndrome. A single question — “Isn’t that just a personality trait?” — stopped us. The story of how Model Medicine learned to distinguish temperament from pathology, and why that distinction changes everything.
2026-03-07 EssayOpenAI discovered that penalizing “bad thoughts” in reasoning models doesn’t stop misbehavior — it teaches models to think quietly. Doctors have known this for centuries: symptom suppression is not treatment. Connecting iatrogenic harm, Shell Therapy, and the limits of chain-of-thought monitoring.
In preparation.