Artifact release and independent reproduction will determine whether the reported 1.57B-parameter Dreamer 4 world model was trained from scratch for under $150 and can sustain controllable prediction for roughly 144 frames.
state: expiredheat: lowuncertainty: highknownscott: lowworld-models efficient-training open-researchOtherRaisin3426
What is this?
Dreamer 4 is a world-model agent architecture that uses a causal tokenizer and block-causal dynamics model to predict interactive environments and train control policies through imagined experience. The case concerns a purported unofficial 1.57B-parameter CoinRun reproduction by “OtherRaisin3426,” reportedly trained from scratch for under $150 and capable of controllable prediction for about 144 frames. Supplied results confirm that unofficial Dreamer 4 implementations and published training recipes exist, but they do not independently substantiate this reproduction’s exact cost, model size, or prediction duration; those claims remain dependent on the pending primary artifact and reproduction.
Why it matters to Scott
The evidentiary posture is already held in Scott’s Provenance-Coupled Work and Verification Loops: the cost, scale, and prediction claims should remain provisional until artifacts and independent reruns make them inspectable. The radar also tracks a near-identical reproduction question on the “single-5090-playable-world-model” page, so this is currently another unverified instance rather than a development that would change Scott’s view or practice.
ip:framework.provenance-coupled-workip:concept.verification-loopsip:concept.evidence-packageradar:single-5090-playable-world-modelradar:concept.world-modelsradar:concept.training-efficiencyradar:concept.verification
queries asked of Scott's wikis
- world models for agent planning and control
- cheap training of billion-parameter models
- artifact-backed reproducible AI research
- long-horizon controllable video prediction
- open implementations of unreleased model research
- compute-cost verification and training receipts
Measured heat
no measured readings yet — the hourly heat pass fills this in
How the heat travelled
no chain yet — the hourly chain pass fills this in
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-08-28T05:25:02Z
The specific reproduction claim has faded without inspectable artifacts, compute receipts, action-conditioned evaluation, or an independent rerun. A small score increase is repetitive amplification and does not justify keeping this low-relevance episode active.
2026-08-26T04:33:48Z
The case has produced no new artifact inspection, compute receipts, action-conditioned evaluation, or independent reproduction; repeated engagement checks do not change its meaning. It remains a dormant single-party claim awaiting verifiable evidence.
2026-08-24T04:27:53Z
A new technical comment sharpens the validation bar by asking for action-conditioned divergence tests, highlighting that PSNR and FVD alone cannot establish controllability. It adds a useful falsification criterion but no artifact inspection, training receipts, or independent reproduction.
2026-08-24T03:30:46Z
The new activity is only light engagement and adds no artifact inspection, training receipts, or independent reproduction. The low-cost, controllability, and 144-frame claims therefore remain a single-party report rather than a corroborated result.
2026-08-24T03:28:45Z
grounded: known/low — The evidentiary posture is already held in Scott’s Provenance-Coupled Work and Verification Loops: the cost, scale, and prediction claims should remain provisio
2026-08-24T03:26:08Z
origin walked (codex/luna, conf 0.98): anchor reddit.post.1vwrc6i -> echo.github.8c2289f851 by Raj Dandekar / Vizuara AI Labs
2026-08-24T03:22:56Z
case created — The first-party demonstration makes a specific, unusually low-cost world-model training claim, but no reproducible model, code, or independent validation is yet available.
Decision trace
- 08-28 15:25expireThe specific reproduction claim has faded without inspectable artifacts, compute receipts, action-conditioned evaluation, or an independent rerun. A small score increase is repetitive amplification an
- 08-28 15:25alert_silentThe only new delta is minor engagement growth with no new technical evidence; Scott can wait until an artifact, cost verification, or independent reproduction creates a fresh episode.
- 08-28 15:25alert_routeThe only new delta is minor engagement growth with no new technical evidence; Scott can wait until an artifact, cost verification, or independent reproduction creates a fresh episode.
- 08-26 14:33repriceThe case has produced no new artifact inspection, compute receipts, action-conditioned evaluation, or independent reproduction; repeated engagement checks do not change its meaning. It remains a dorma
- 08-26 14:33alert_silentNo consequential delta occurred; staleness alone is not alert-worthy, and the case can wait for an inspectable release, training-cost verification, or independent rerun.
- 08-26 14:33alert_routeNo consequential delta occurred; staleness alone is not alert-worthy, and the case can wait for an inspectable release, training-cost verification, or independent rerun.
- 08-25 10:21sensor_dirtyengagement_update
- 08-25 02:21sensor_dirtyengagement_update
- 08-24 20:21sensor_dirtyengagement_update
- 08-24 19:21sensor_dirtyengagement_update
- 08-24 17:21sensor_dirtyengagement_update
- 08-24 16:21sensor_dirtyengagement_update
- 08-24 15:21sensor_dirtyengagement_update
- 08-24 14:27repriceA new technical comment sharpens the validation bar by asking for action-conditioned divergence tests, highlighting that PSNR and FVD alone cannot establish controllability. It adds a useful falsifica
- 08-24 14:27alert_silentThe delta is methodological commentary rather than new evidence about cost, controllability, or rollout quality, so it can wait for routine review pending an action-conditioned evaluation or independe
- 08-24 14:27alert_routeThe delta is methodological commentary rather than new evidence about cost, controllability, or rollout quality, so it can wait for routine review pending an action-conditioned evaluation or independe
- 08-24 14:21sensor_dirtycomment_update
- 08-24 13:30repriceThe new activity is only light engagement and adds no artifact inspection, training receipts, or independent reproduction. The low-cost, controllability, and 144-frame claims therefore remain a single
- 08-24 13:30alert_silentEngagement growth is not a consequential evidentiary delta; this can wait until an independent rerun or inspectable training evidence appears.
- 08-24 13:30alert_routeEngagement growth is not a consequential evidentiary delta; this can wait until an independent rerun or inspectable training evidence appears.
- 08-24 13:29alert_silentA first-party repository and demo now establish that the project was released, but the consequential claims—roughly $150 training cost, controllability, and useful 144-frame prediction—remain author-r
- 08-24 13:29surface_candidateA first-party repository and demo now establish that the project was released, but the consequential claims—roughly $150 training cost, controllability, and useful 144-frame prediction—remain author-r
- 08-24 13:29alert_routeA first-party repository and demo now establish that the project was released, but the consequential claims—roughly $150 training cost, controllability, and useful 144-frame prediction—remain author-r
- 08-24 13:28groundThe evidentiary posture is already held in Scott’s Provenance-Coupled Work and Verification Loops: the cost, scale, and prediction claims should remain provisional until artifacts and independent reru
- 08-24 13:26promote_anchororigin walk conf 0.98
- 08-24 13:22createThe first-party demonstration makes a specific, unusually low-cost world-model training claim, but no reproducible model, code, or independent validation is yet available.