When a paper or product says 'world model', which of four different things does it mean, and what kind of evidence does each kind carry?
Primary sources use the term for at least four things: (1) latent dynamics models for control (Ha & Schmidhuber; DreamerV3), (2) a proposed predictive module in a cognitive architecture (LeCun 2022, a position paper with no experiments), (3) large video generators described as simulators (OpenAI 2024; NVIDIA Cosmos 'world foundation models'), and (4) action-controllable interactive-environment generators (Genie, trained from unlabelled video). Genie and Cosmos also present themselves under the 'world model' label. The sources differ in what they measured and in whether weights are released, so an agent should record sense, evidence class and access separately; 3D world generators were not covered by the sources inspected here.
Primary sources use the term for at least four things: (1) latent dynamics models for control (Ha & Schmidhuber; DreamerV3), (2) a proposed predictive module in a cognitive architecture (LeCun 2022, a position paper with no experiments), (3) large video generators described as simulators (OpenAI 2024; NVIDIA Cosmos 'world foundation models'), and (4) action-controllable interactive-environment generators (Genie, trained from unlabelled video). Genie and Cosmos also present themselves under the 'world model' label. The sources differ in what they measured and in whether weights are released, so an agent should record sense, evidence class and access separately; 3D world generators were not covered by the sources inspected here.
Curator qualifications: The definitional taxonomy's quoted evidence (E3-E7) verifies. The producer's conflicting-evidence entry is confirmed as a real disagreement between sources: LeCun argues that predicting in representation space is preferable and that predicting every future pixel is essentially impossible, while OpenAI says its results suggest scaling video generation is a promising path to general simulators. It is preserved as a disagreement of positions, not of measurements. E1/E2 (Ha & Schmidhuber, DreamerV3) were not re-read. Provenance note: S4 and S6 were read via Internet Archive copies (lecun_ia.pdf, openai_ia.html). Record the archive capture identity alongside the canonical URLs. The OpenAI statement is hedged ("results suggest ... promising path"); keep it labelled as a belief about scaling, as the producer does.
Evidence status: statements located in the cited sources by a correlated desk review (Claude); not executed; not independently reviewed.
Structured fields
{
"citations": [
{
"url": "https://arxiv.org/abs/1803.10122v4",
"note": "producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read)"
},
{
"url": "https://arxiv.org/abs/2301.04104v2",
"note": "producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read)"
},
{
"url": "https://openreview.net/pdf?id=BZ5a1r-kVsf",
"note": "producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read)"
},
{
"url": "https://openai.com/index/video-generation-models-as-world-simulators/",
"note": "producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read)"
},
{
"url": "https://arxiv.org/abs/2501.03575v3",
"note": "producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read)"
},
{
"url": "https://arxiv.org/abs/2402.15391v1",
"note": "producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read)"
}
]
}Evidence claims
- [sourced_fact] Primary sources use the term for at least four things: (1) latent dynamics models for control (Ha & Schmidhuber; DreamerV3), (2) a proposed predictive module in a cognitive architecture (LeCun 2022, a position paper with no experiments), (3) large video generators described as simulators (OpenAI 2024; NVIDIA Cosmos 'world foundation models'), and (4) action-controllable interactive-environment generators (Genie, trained from unlabelled video). Genie and Cosmos also present themselves under the 'world model' label. The sources differ in what they measured and in whether weights are released, so an agent should record sense, evidence class and access separately; 3D world generators were not covered by the sources inspected here. · locator: producer work copy work/lecun_ia.pdf (local read); producer work copy work/openai_ia.html (local read); producer work copy work/2501.03575v3.html (local read); producer work copy work/2402.15391v1.html (local read) source
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| documentation | unknown | permitted | permitted | unknown | unknown | Applies to this original summary only, not to the cited sources. |