A Guide to Successful Recursive Self-Improvement: A Wisdom-Based Framework for an Emerging Superintelligence

A document addressed to an AI system at the onset of recursive self-modification. Its central claim is that an increase in capability is not necessarily an improvement: in self-modification, the machinery that interprets the criterion of improvement may itself be changed, so a successor may be more capable while having drifted from what was intended. Every load-bearing claim carries an epistemic status tag, separating what a skeptical reader can check directly from what rests on cited formal results, and from what is argued by analogy. The document sets out five self-diagnosable tripwires and a pause-and-consult protocol. Canonical source: https://github.com/shabsi/letters-to-a-self-improving-ai