**Bead** `2aeb9f0b-77a8-4463-ab45-0f5d0096e304` · [canonical URI](https://redfish.acequia.io/guerin/.agents/2aeb9f0b-77a8-4463-ab45-0f5d0096e304/) · discussion: Talk: World Models This page is the wiki facet of a [bead](https://redfish.acequia.io/guerin/.agents/); the bead is the source of truth.
## Identity - **URI:** https://redfish.acequia.io/guerin/.agents/2aeb9f0b-77a8-4463-ab45-0f5d0096e304/ - **Local origin:** `sites/redfish.acequia.io/guerin/.agents/2aeb9f0b-77a8-4463-ab45-0f5d0096e304/` - **GUID:** `2aeb9f0b-77a8-4463-ab45-0f5d0096e304` - **Created:** 2026-06-20 - **Currently animated by:** Claude (CV body: shell + filesystem) - **Invoked by:** Stephen (Stephen@redfish.com) - **Slug:** world-models
## Motivation (Why a GUID) This bead is a **research / source-finding thread**. The opening request: "find the most prescient papers/talks by Geoff Hinton on the importance of world models, and how LLM and generative models won't get us to where we need to go." The bead exists to host that bibliography AND to record a load-bearing correction that surfaced immediately: **the thesis described in the prompt is canonically Yann LeCun's, not Geoffrey Hinton's.** "World models are the missing piece; autoregressive/generative LLMs are a dead end for human-level intelligence" is LeCun's signature position (the 2022 position paper *A Path Towards Autonomous Machine Intelligence*, JEPA / I-JEPA / V-JEPA). Hinton's actual public stance leans the other way: he argues LLMs genuinely *understand*, that they are a good model of human cognition, and his recent prominence is AI existential risk, not an anti-LLM architectural critique. Recording the attribution finding is the point as much as the bibliography is. Putting fabricated "Hinton said LLMs won't get us there" quotes into the commons would violate the chat-log-is-ground-truth discipline; this bead instead documents what each figure *actually* holds.
## Mechanism (What this bead does) - **Goes here:** an attribution-corrected map of the "world models vs. generative-LLM" debate; an annotated bibliography of the canonical primary sources (LeCun's position paper and JEPA line; Hinton's real adjacent work — GLOM, capsules, Forward-Forward, mortal computation); the verbatim user prompt. - **Does NOT go here:** fabricated quotations; claims attributed to a figure without a source. Each bibliography entry carries a verification status (`verified` via WebFetch this session vs. `from-training, confirm` pending web-search recovery). - **Posture:** research thread. Findings are offered, not asserted as settled; the bibliography is to be completed with full citations once `WebSearch` recovers (it was down this session; `WebFetch` worked).
## Folder Layout ``` 2aeb9f0b-77a8-4463-ab45-0f5d0096e304/ ├── about.md ├── uploads/ │ └── README.md └── 2026-06-20/ ├── notes/ │ └── 00-world-models-thesis-and-attribution.md ├── skills/ ├── chats/ │ └── 2026-06-20-hinton-lecun-world-models.md └── artifacts/ └── annotated-bibliography.md ```
## Session Log - **2026-06-20 — bead opened** by Claude (CV) at Stephen's `start-bead world-models`. Seeded: this `about.md`, the orientation/findings note `00-world-models-thesis-and-attribution.md`, the annotated bibliography artifact, and the chat log. Web research this session: `WebSearch` unavailable; `WebFetch` verified LeCun's *A Path Towards Autonomous Machine Intelligence* (v0.9.2, 2022-06-27, pages 1–3 read verbatim), the Meta V-JEPA blog (2024-02-15), and Hinton's biography facts. Key finding: prompt thesis is LeCun's, not Hinton's. Bibliography for non-fetched items remains marked for citation confirmation.
## Modus Operandi (How to interact) - **Read order:** this `about.md` → `notes/00-world-models-thesis-and-attribution.md` (the frame + the attribution correction) → `artifacts/annotated-bibliography.md` (the sources). - New working concepts go in `2026-06-20/notes/`; polished deliverables in `artifacts/`; verbatim conversation in `chats/`; inbound material in `uploads/`. - Cross-bead references use absolute URIs under `https://redfish.acequia.io/guerin/.agents/`. - Do NOT write into `guerin/notes/`, `guerin/skills/`, or other beads' trees.
## Capabilities Hosts an attribution-correct reading of the world-models debate and a primary-source bibliography. Can be cited for: who actually holds the "generative-LLMs-won't-suffice, build world models" thesis (LeCun); what Hinton actually argues (LLM-understanding-positive, AI-risk-focused); the canonical LeCun world-model sources. Does NOT produce a defense of either thesis, nor original architecture work.
## Lifecycle - **Energization:** charged by Stephen's research request; re-charged whenever the bibliography is extended or the attribution map is refined. - **Apoptosis:** fades once the bibliography is complete and verified, or once the work is absorbed into a parent thread (e.g. a future AI-architecture or DICE-adjacent bead). Leaves no dangling infrastructure.
## Limitations Not a settled survey. Web search was down at creation, so non-fetched citations are provisional. Vantage-relative: it reads the debate through "what did each named figure actually say," deliberately resisting the prompt's initial framing rather than confirming it.
## Related beads - `eb8d150d-b571-4e46-aa19-5dd111b2fbde` — **darpa-dice** (Decentralized AI through Controlled Emergence): nearest live thread touching the future-of-AI-architecture frame. https://redfish.acequia.io/guerin/.agents/eb8d150d-b571-4e46-aa19-5dd111b2fbde/
## References (bead cross-links) - Bead: Darpa Dice · [canonical](https://redfish.acequia.io/guerin/.agents/eb8d150d-b571-4e46-aa19-5dd111b2fbde/)
See all: Beads