From Personal LLMs to Globally Connected AI: The Need for a Knowledge World with Memory as Software
AbstractContemporary large language models (LLMs) substantially accelerate intellectual activities organized around individual conversations, files, tasks, and workspaces. Yet research and knowledge generation do not take place entirely within individuals. Questions, branching hypotheses, verification and refutation, failures, discovery, organization, and the production of outputs develop through relationships spanning multiple participants, topics, and histories. Even as personal LLMs become more capable, their contributions may remain local if they are weakly connected to these processes of knowledge generation.This paper proposes constructing a knowledge world that preserves the relationships and histories of knowledge generation, together with globally connected LLMs that operate across that world and globally connected AI systems that incorporate them. The central design proposition is that memory is software. Records of the past do more than remain stored: they change which questions are explored next, which hypotheses are reconsidered, which failures are avoided, which people and studies are connected, and which outputs are produced. What matters is an operational structure through which memory influences future intellectual actions.After defining globally connected LLMs and AI and describing their architecture and operational cycle, the paper examines the strengths and limits of personal LLMs and the relation between memory software and the knowledge world. As a candidate implementation, it discusses the MMO Research Platform, with its research islands, discoverers, diverse participant roles, permanent-conference structure, and LLM-assisted moderation. Visibility, consent, access rights, attribution of contributions, and public-interest governance are treated as design requirements. The paper thereby extends the agenda for LLM development beyond model performance to the construction of the memory worlds in which models operate. Keywordspersonal LLMs; globally connected LLMs; globally connected AI; memory as software; knowledge world; collective intelligence; MMO Research Platform; discoverers; LLM UX AI Assistance DisclosureGenerative AI was used as an auxiliary tool in drafting, adjusting expressions, and organizing the structure of this paper. The author bears final responsibility for the text, claims, citations, and factual verification.
Authors
- Kei Itoh (ORCID: https://orcid.org/0009-0008-7370-1365)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24274338
- Primary Topic
- Artificial Intelligence Applications
- Type
- preprint