Qu, Yansong, Chen, Jiancong, Labi, Samuel · Nov 3, 2026
Jev with Foresight: Safe Language-Guided Autonomous Driving via a Language World Model
Markdown · JSON · Original source
Authors Yansong Qu, Jiancong Chen, and Samuel Labi present -AD, which pairs Jev—a typed decision model that answers predefined questions with probabilities—with a language that predicts how nearby vehicles move under each candidate action. Code converts those predictions into foreseen consequences for Jev to compare, rather than requiring it to infer consequences from coordinates. In simulated highway merging, the authors report that foresight lowers the same decision model’s collision rate from 26.0% to 2.4% at 0.25 seconds per decision; Jev-AD achieves passengers’ intent in 93% of free-form command trials. In 500 replayed real-world NGSIM episodes, they report a decrease from 48.2% to 2.8%, and say model foresees 93% of dangerous actions. These results suggest a way to make language-guided driving decisions by comparing predicted consequences, but the reported evaluations are simulations and replayed episodes, not evidence of performance in deployed vehicles.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 3, 2026 (2026-11-03).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T09:33:40.070Z. First observed: 2026-11-05T05:00:25.040Z. Retrieved: 2026-11-09T09:32:45.956Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The abstract reports results from simulated highway merging and replayed real-world NGSIM episodes; it does not report evaluation in deployed vehicles. The evidence provided is the paper’s page and abstract, not the full paper.
Open question
How would -AD perform beyond the simulated highway-merging trials and replayed NGSIM episodes reported by the authors?
Source-stated
The authors present Jev-AD, combining Jev, a typed decision model that answers predefined questions with probabilities, and a language world model that predicts nearby vehicles’ movement for candidate actions.
Jev, a typed decision model that answers predefined questions with probabilities, not generated text, with a language world model that predicts for every candidate action how nearby vehicles move over the next two seconds.
Source-stated
The authors say code turns the predictions into foreseen consequences for Jev to compare.
Code turns these predictions into foreseen consequences, which Jev compares instead of inferring them from coordinates.
Source-stated
In simulated highway merging, the authors report that foresight reduced the same decision model’s collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and that Jev-AD achieved passenger intent in 93% of free-form command trials.
In simulated highway merging (500 car-episodes), this foresight lowers the same decision model's collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and Jev-AD achieves the passenger's intent in 93% of free-form command trials.
Source-stated
For 500 replayed real-world NGSIM episodes, the authors report a collision-rate decrease from 48.2% to 2.8%, and say the world model foresaw 93% of dangerous actions.
In 500 replayed real-world NGSIM episodes, foresight lowers the collision rate from 48.2% to 2.8%, and the world model, learned from this traffic, foresees 93% of dangerous actions.
Escape Meridian interpretation
The reported evaluations are simulations and replayed episodes, so these results alone do not establish performance in deployed vehicles.
In simulated highway merging (500 car-episodes)
Source and contributor context
Qu, Yansong, Chen, Jiancong, Labi, Samuel · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
World model
An AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
Jev
Jev is TypeSafe AI’s model for fast evaluation and routing decisions inside AI systems.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
The World
The World is a public radio news program publishing international reporting on politics, society, labor and technology.
Editorial provenance
Version 1. Publication ID: a77ee978-d673-4ffe-8288-83a2640f2469.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T09:33:40.070Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 0c6228cd11917ef857a7dfe97d9d7c60135950ffdf25d0d5fc2f00934129a456. Source text hash: cd462862040fe9606def6c50b8868f163ec575c1bb2b997f2e8659c30b19e805.
Revision and source-check history
Version 1 · 2026-11-09T09:33:40.070Z
Jev with Foresight: Safe Language-Guided Autonomous Driving via a Language World Model
Authors Yansong Qu, Jiancong Chen, and Samuel Labi present -AD, which pairs Jev—a typed decision model that answers predefined questions with probabilities—with a language that predicts how nearby vehicles move under each candidate action. Code converts those predictions into foreseen consequences for Jev to compare, rather than requiring it to infer consequences from coordinates. In simulated highway merging, the authors report that foresight lowers the same decision model’s collision rate from 26.0% to 2.4% at 0.25 seconds per decision; Jev-AD achieves passengers’ intent in 93% of free-form command trials. In 500 replayed real-world NGSIM episodes, they report a decrease from 48.2% to 2.8%, and say model foresees 93% of dangerous actions. These results suggest a way to make language-guided driving decisions by comparing predicted consequences, but the reported evaluations are simulations and replayed episodes, not evidence of performance in deployed vehicles.
Original source
cs.AI · Article
Source details
Original source date: Nov 3, 2026 (2026-11-03).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T09:33:40.070Z. First observed: 2026-11-05T05:00:25.040Z. Retrieved: 2026-11-09T09:32:45.956Z.
Limits
The abstract reports results from simulated highway merging and replayed real-world NGSIM episodes; it does not report evaluation in deployed vehicles. The evidence provided is the paper’s page and abstract, not the full paper.
Open question
How would -AD perform beyond the simulated highway-merging trials and replayed NGSIM episodes reported by the authors?
Source-stated
The authors present Jev-AD, combining Jev, a typed decision model that answers predefined questions with probabilities, and a language world model that predicts nearby vehicles’ movement for candidate actions.
Jev, a typed decision model that answers predefined questions with probabilities, not generated text, with a language world model that predicts for every candidate action how nearby vehicles move over the next two seconds.
Source-stated
The authors say code turns the predictions into foreseen consequences for Jev to compare.
Code turns these predictions into foreseen consequences, which Jev compares instead of inferring them from coordinates.
Source-stated
In simulated highway merging, the authors report that foresight reduced the same decision model’s collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and that Jev-AD achieved passenger intent in 93% of free-form command trials.
In simulated highway merging (500 car-episodes), this foresight lowers the same decision model's collision rate from 26.0% to 2.4% at 0.25 seconds per decision, and Jev-AD achieves the passenger's intent in 93% of free-form command trials.
Source-stated
For 500 replayed real-world NGSIM episodes, the authors report a collision-rate decrease from 48.2% to 2.8%, and say the world model foresaw 93% of dangerous actions.
In 500 replayed real-world NGSIM episodes, foresight lowers the collision rate from 48.2% to 2.8%, and the world model, learned from this traffic, foresees 93% of dangerous actions.
Escape Meridian interpretation
The reported evaluations are simulations and replayed episodes, so these results alone do not establish performance in deployed vehicles.
In simulated highway merging (500 car-episodes)
Source and contributor context
Qu, Yansong, Chen, Jiancong, Labi, Samuel · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
World model
An AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
Jev
Jev is TypeSafe AI’s model for fast evaluation and routing decisions inside AI systems.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
The World
The World is a public radio news program publishing international reporting on politics, society, labor and technology.
Editorial provenance
Version 1. Publication ID: a77ee978-d673-4ffe-8288-83a2640f2469.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T09:33:40.070Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 0c6228cd11917ef857a7dfe97d9d7c60135950ffdf25d0d5fc2f00934129a456. Source text hash: cd462862040fe9606def6c50b8868f163ec575c1bb2b997f2e8659c30b19e805.
uncertain · 2026-11-09T09:33:40.070Z · anthropic/claude-opus-5.5
Publication: a77ee978-d673-4ffe-8288-83a2640f2469
attribution — pass: The reading correctly credits the arXiv authors Yansong Qu, Jiancong Chen and Samuel Labi. It keeps the performance results attributed to them ('the authors report', 'say') rather than presenting them as established facts.
Authors:Yansong Qu, Jiancong Chen, Samuel Labi
summary — uncertain: The numbers are faithful, but the reading drops the problem the authors set up. In the abstract, large language models used as drivers reason in generated text for seconds and infer consequences implicitly. Without that, a general reader cannot see why '0.25 seconds per decision' or 'probabilities, not generated text' matters. The reading also omits the two-second prediction window that defines 'foresight'. Smallest repair: add one opening sentence stating that motivation and restore 'over the next two seconds'. 'Typed decision model' and NGSIM also need term-library tooltips.
as driving agents they reason in generated text for seconds and must infer each action's consequences implicitly from the scene
date — pass: The displayed original publication date of 2026-11-03 matches the v1 submission date on the arXiv page. The later feed date of 2026-11-05 is correctly not used.
[Submitted on 3 Nov 2026]
support — pass: Each figure is quoted accurately from the abstract: 26.0% to 2.4%, 0.25 s per decision, 93% intent, 48.2% to 2.8%, and 93% of dangerous actions foreseen. The deployment caveat is labeled as interpretation and stays within the evidence. The question follows from the stated evaluation scope rather than leading the reader.
In 500 replayed real-world NGSIM episodes, foresight lowers the collision rate from 48.2% to 2.8%
media — pass: The selected media is the verified direct arXiv abstract page for this paper. That is the most direct way for a reader to reach the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
Jev is TypeSafe AI’s model for fast evaluation and routing decisions inside AI systems.
Explore JevAn AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
Definition referenceThe World is a public radio news program publishing international reporting on politics, society, labor and technology.
Explore The WorldarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivJev is TypeSafe AI’s model for fast evaluation and routing decisions inside AI systems.
Explore JevJev is TypeSafe AI’s model for fast evaluation and routing decisions inside AI systems.
Explore JevAn AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
Definition referenceThe World is a public radio news program publishing international reporting on politics, society, labor and technology.
Explore The WorldarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivJev is TypeSafe AI’s model for fast evaluation and routing decisions inside AI systems.
Explore Jev