Informative—thanks for sharing. What specific patterns can be observed in the behavioral scheming AI system post-training? What steps do/could/should the human participate employ as a response?