Personal World Models

Memory is a compression problem. A life compressed the way a map compresses territory: built once, read by any model, owned by the person.

Memory is a compression problem. Current systems treat it as retrieval, which can say what happened but not what it means or what comes next. We introduce the Personal World Model (PWM): a life compressed the way a map compresses territory: what recurs becomes structure, and any model can read the structure and anticipate from it. GOLGI instantiates the framework: it ingests a real four-week photo archive, 720K tokens of raw life, into anchored entities, routines, and narratives an agent traverses in hundreds of tokens, on a pipeline that runs end-to-end on 4B on-device models. The agent…

saved by

• Carmen Gutierrez

related reading

• Verbalizable Representations Form a Global Workspace in Language Modelstransformer-circuits.pub
• The Persona Selection Model: Why AI Assistants might Behave like Humansalignment.anthropic.com
• Guardian Angels: LLM Personalization for Productivity and Security · Gwern.netgwern.net
• Co-Creating Life Goals from Everyday Computer Usearxiv.org
• Memory Models: Towards Agents That Learnletta.com
• World Models: Computing the Uncomputablenotboring.co
• [2505.10831] Creating General User Models from Computer Usearxiv.org
• Engramme – Remember Everythingengramme.com
• [2605.15156] MeMo: Memory as a Modelarxiv.org
• The Continual Learning Problemjessylin.com
• pdfopenreview.net
• LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixelsle-wm.github.io