SIG-PSY 2026-08-13

Session Recording Summary · 55m 34s · Full notes ↗

Reading: Not clearly identified in the discussion. This session was not a reading-discussion in the usual sense; it was a demo and working session for the group's "World Machines" project tools (the Oracle, the Prime Radiant, the Scribe note-taking app, and related infrastructure). Aneesh also mentioned a piece he "published last night" outlining principles used in the Prime Radiant, which he shared in chat, but no specific shared reading was worked through.

Aneesh gave an extended walkthrough of the software he has been building for the group: a note-taking/wiki app ("Scribe"), the "Oracle" retrieval system built on the group's corpus of books and essays, and the "Prime Radiant" — a temporal knowledge-graph system that models history as interacting "machines" and generates extrapolations/"prophecies." The second half turned to planning the upcoming symposium/special session (involving Aneesh, Sean, and Florian), which Venkatesh framed as the project's public launch. Discussion touched on how notes and essays could eventually feed automatically into the modeling system, and on how contributions and funding might scale afterward.

• **Project context (Aneesh):** The group began as a book club on Venkatesh's Substack, evolved into members writing and cross-referencing essays, and became the "World Machines" project. The core framing (attributed to Venkatesh) is that history can be read as different "machines" in play — e.g., a modernity machine, a "versions" machine, a "liveness" machine.
• **The Oracle:** An AI system that ingests the group's material (books, members' essays, external essays) into a searchable database so members can ask questions or explore, ideally lowering the barrier to writing essays. It functions as retrieval-and-answer ("it doesn't do prophecy yet"), and behind the scenes searches a **knowledge graph**, taking several back-and-forth steps across related links before answering.
• **Two-tier vs. MCP access (Aneesh):** The portal AI is described as "fairly low powered" and uses a two-tier agent setup (the Oracle queries a second agent with access to raw notes, which returns summaries). A dedicated **MCP** option lets members connect their own, more capable AI directly to the corpus (including the raw text) via a generated access token.
• **Scribe app (Aneesh):** `scribe.worldmachines.org` — a lightweight web app (installable as a PWA on iPhone via "Add to Home Screen") for quickly adding notes to the corpus. Motivation: editing/uploading text files was too much friction and meant he wasn't writing anything. Notes can be quick thoughts, book notes, or "velocity notes" / term definitions (Venkatesh had wanted to define some terms).
• **Notes as corpus / "hive mind" (Aneesh):** The purpose of adding a note is that it becomes a source the Oracle can cite. Aneesh described it as a hive mind where members' thoughts enter the shared sourcing and others (and their AIs) can build on them.
• **Provenance/tagging (Aneesh):** There's no truth filter, but every note is attributed to a person or tagged as coming from a book — so a book source may be more trustworthy than a personal opinion.

Questions & Disagreements: - **Ellen** asked clarifying questions as a newcomer: what the purpose of adding a note is, whether it's just to continue book-club discussion, and whether there's a human-readable view or if the corpus is only AI-accessible — concluding "it's a wiki." - **Spencer** asked how the system handles **factual differences vs. perspective differences** in notes (e.g., if two people believe different thin

Participants: Aneesh Sathe, Sean Stevenson, Spencer Norwick U-0700, Andre Comeau, Ellen Kaye-Cheveldayoff (UTC -4), Kyle Mathews, Florian Lohse (UTC+0), rafa (UTC+1), Venkatesh Rao