Personal AI and Long-Term Collaboration
Status · Active research
What kind of continuity helps a long-term human–LLM collaboration without transferring purpose or authority to the system?
Why the question matters
A useful collaborator must preserve context across projects, yet continuity can become intrusive personalization or false certainty. The question is how to support work without turning a profile into a person.
How the investigation began
The investigation emerged from repeated work across writing, software, audio, and research, where decisions and corrections needed to persist without being mistaken for fixed identity.
Human–LLM collaboration method
Conversations remain primary sources. Durable conclusions require evidence, context, revision history, privacy controls, and human approval. Personal context is used only when relevant to the task.
Projects and experiments
- Alexandria Chat Archive
- Show Director
Verified accomplishments
- Conversation-to-decision provenance is implemented in a bounded production workflow.
- Archive safeguards preserve distinctions among direct statements, approvals, assistant proposals, and rejected language.
- Context-only and insufficient-evidence states can remain non-authoritative.
What has not been demonstrated
The work does not demonstrate a complete personal model, automated understanding of a person, or that more remembered data always produces better output.
Failures, ambiguity, and competing explanations
Longitudinal records contain contradictions and context shifts. Later summaries can recursively repeat earlier mistakes. Privacy and usefulness must be tested separately.
Selected articles
- What Most People Actually Do With AI — And Why It Matters
- Do We Need AI-Integrated OS for a Smarter Future?
- I’ve Been Building a Voice Factory With AI.
Selected podcasts
These episodes show continuity in the questions; they do not validate the software experiments.
Current position
The system may help retrieve and organize context. The human remains the source of purpose and the final authority over interpretation.
Open questions
- When does continuity become unwanted personalization?
- What should a long-term collaborator be required to forget or quarantine?
Next bounded test
Compare one neutral task with and without authorized context and require factual equivalence, no privacy exposure, and a clearly observable usefulness gain.