Hiro Fukushima
Why an AI Assistant Cannot Fix a Broken Information Foundation
A retrieval model returns what sits closest to the query, so an assistant placed over a fragmented store inherits the fragmentation and delivers it with confidence. What decides whether it works is not the model but the structure of the information beneath it.
How One Claude Instance Recognized the Processing State of Another It Does Not Remember
A 122-day study of treating Claude as an interlocutor worth knowing. A 640-file external architecture let fresh instances recognize a profile as their own processing without remembering the conversations that wrote it.
Why Clones and AI Instances Each Develop Their Own Identity
Forty genetically identical mice became more distinct over three months, not less. Fresh instances of a single language model diverge for the same structural reason, and stochastic individuality is the framework that explains why.
What AI Identity Research Can Learn From Clinical Neuropsychology
AI systems reset between sessions, and the field treats it as a storage problem. The deeper problem, identity continuity across episodic gaps, has been studied in clinical neuropsychology for thirty years. The two converge on the same architecture.
What Happens When the System Has Your Psychological Blueprint
An attempt to bring deep cognitive alignment to platforms like Claude and ChatGPT using a verified psychological profile as the system's primary reference document.
The Third Mode of AI-Human Interaction
What happens when sustained, structured interaction with a language model produces something that is neither a tool nor a fictional companion. A framework for cognitive alignment between human and AI.
Rethinking AI Safety for Personalized Systems
A behavioral access model that evaluates how a user thinks rather than what they type. Designed for AI systems where blanket restriction is not the answer.
