Large Universe Models for personal use — a companion that stays current

What it ingests

Calendar, correspondence, project and task state, documents in progress, health and training data, finances, reading and browsing history — whichever of these you choose to connect, and nothing else.

Why continuity matters for one person

The gap between a personal assistant and a colleague is not intelligence. It is that one of them has been there all year.

A Large Language Model can help with whatever you paste into it. It has no idea what you decided in March, what you have been avoiding since June, or that the assumption behind your current plan stopped being true last week. Each conversation begins from nothing.

A Large Universe Model holds the state of your commitments between conversations, and revises it as evidence arrives.

The sequence is always the same. A Large Language Model read a corpus once and stopped. A Large World Model learned to simulate a scene it was shown. A Large Universe Model keeps watching, and revises.

Honest project state

Everyone has projects that are formally active and actually stalled. The gap is uncomfortable and therefore rarely examined.

A Large Universe Model tracks the last real activity on each, holds a belief about whether it is progressing, and can tell you — without judgement and without being asked — that four of your eleven open projects have had no substantive movement in six weeks.

Reconciliation

What you committed to, against what your calendar and inbox show actually happening. Not as a productivity score, but as a factual mismatch worth noticing: you agreed to three things this month that no time has been allocated to.

Revisiting decisions

The hardest thing to do unaided is return to a decision when its reasoning stops holding. You made a choice in March for stated reasons. Two of those reasons are no longer true, and nothing in your life is structured to tell you.

A Large Universe Model that recorded the reasoning as a belief with dependencies raises it when the dependencies move. This is the single most valuable personal application, and it is impossible without persistence.

Speaking rarely

The design constraint is restraint. A companion that comments continuously is worse than none. The trigger should be a confident belief breaking — not activity, not thresholds, not a daily digest.

Privacy

This application involves the most sensitive data a person has. Any honest treatment has to say that continuous ingestion of a life is a serious thing to hand to a system, and that where it runs and who can read the belief state matter more than any capability described above.

Large Universe Models for individuals