RoamDemoJuly18

RoamDemoJuly18 centers on a plan to build “Rome”: a peer-to-peer map of the world’s knowledge that helps people find efficient routes through information toward the research frontier. The core claim is that such a map can’t be assembled from static, interconnected documents alone; it requires groups of people to first model their own problem spaces. That leads to a prerequisite tool—called a “personal Memex”—designed to let individuals flexibly organize concepts as they think, then reuse those organized pieces across many contexts.

The demo shows how the personal Memex works as an evolving knowledge workspace rather than a single document repository. One key feature is zooming between levels of abstraction, similar to outliners, so a user can move from a high-level view down to specific notes and back again. Another central capability is “mentions,” which allow the same idea to be referenced in multiple locations. Instead of duplicating content, the system tracks where an idea appears and lets users navigate across those references.

A concrete example uses a paper titled “Concrete Problems in AI Safety.” The workflow begins with document analysis: the user annotates the paper, applies filters to carve away questionable sections, and then navigates directly to relevant parts of the text. Those annotated sections can then be dragged into an outline, where they become reusable building blocks—referenced not only in the original document but also in other structures. The demo emphasizes that this supports charting a path through research papers while still allowing the user to reorganize and reuse insights as understanding changes.

The organizing philosophy is framed as “carving reality at the joints,” borrowing a quote about taking in scattered particulars under one idea, then separating the idea into parts “as nature directs” without breaking anything important. In practice, that means the system should help users aggregate potentially relevant information even when they don’t yet know the correct structure. The personal Memex therefore focuses on flexible organization: explicitly building a graph of concepts, surfacing related items, and letting users filter what’s relevant to a current question.

Graph-based organization is presented as a core feature. As the user works through an idea, they can zoom in on relevant relationships—such as items related to categories or items tied to a specific concept. The demo also illustrates how the same concept can be labeled differently depending on the user’s longer-term framing, including “single player mode” and “multiplayer mode,” with “personal Memex” and “dynamic knowledge repository” used as related labels in that roadmap.

Finally, the demo situates the tool in a larger research collaboration goal. Work with researchers aims to map problem spaces and understand what research teams need to collaborate more effectively, as the system moves toward a multiplayer mode. Feedback is solicited on the approach and features, particularly around how well the tool supports evolving, graph-based knowledge organization.