Papers
A real question was put to AI Macx. Here is the answer we were willing to publish.
Each paper takes a question we were actually handed by a client, and publishes the reasoning instead of the resume. The client, the data, the economics, and the implementation stay out. The architecture, the diagnosis, and the things that went wrong stay in ... including the ones still unresolved. Every paper carries a section on what failed, because a method that has only ever worked has not been run.
On what is withheld
These papers are composites. They draw on more than one engagement, and no single client is described. Details that would identify a client have been changed or removed; details that would reveal implementation have been withheld. Where an outcome is stated, it is directional and drawn from work we actually did. Where we do not know something, we say so instead of redact it.
The series continues on both sides of the house. Paul Ann Labs publishes the operations side ... what an AI system inside a company is allowed to believe, and allowed to do.