Shi, Danqing, Jiang, Lan, Lyons, Jake, Zilka, · Oct 28, 2026
Researchers propose eye tracking to assess AI-generated video realism
Markdown · JSON · Original source
Danqing Shi, Lan Jiang, Jake Lyons and Zilka argue that ratings of AI-generated videos capture viewers’ conclusions but not the perception leading to them. They study eye tracking as a complementary signal, recording where viewers look while watching videos. In two Visual Turing Test experiments, 90 participants watched videos generated by recent models twice and then gave feedback. The researchers report 4,286 valid eye-tracking trials across 438 videos, and say gaze patterns align with viewers’ explicit feedback. They introduce GazeEval, a computational method that predicts perceived visual realism from eye-tracking data. Its reported prediction accuracy is 71.3%; the authors present this as evidence that eye-tracking data carries information about how people evaluate a video. They release the study data for further research. The abstract does not establish how well the method would work with other viewers, videos or viewing conditions.
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: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T09:08:11.790Z. First observed: 2026-10-30T06:43:01.472Z. Retrieved: 2026-10-31T06:32:57.905Z.
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 available evidence is the paper’s abstract; it does not give details about participant characteristics, the videos or models tested, or how prediction accuracy was calculated. The abstract reports prediction accuracy and alignment with explicit feedback, but does not establish performance beyond the experiments described.
Open question
How well does GazeEval’s prediction of perceived visual realism hold beyond the 438 videos and 90 participants in these experiments?
Source-stated
The authors say explicit ratings capture viewers’ conclusions but not the perception behind them, and investigate eye tracking as a complementary signal.
While these signals capture the conclusions viewers reach, they fail to offer insights into the perception that leads to them.
Source-stated
Across two experiments, 90 participants watched videos twice and provided feedback; the researchers collected 4,286 valid eye-tracking trials on 438 videos.
Across the experiments, we collected 4,286 valid eye-tracking trials on 438 videos.
Source-stated
The authors report that gaze patterns provide informative evaluation signals that align with explicit human feedback.
Our analysis indicates that gaze patterns offer informative evaluation signals that align with explicit human feedback.
Source-stated
The authors propose GazeEval to predict perceived visual realism from eye-tracking data and report 71.3% prediction accuracy.
It achieves 71.3% accuracy in prediction, providing evidence that eye tracking data carries informative signals about how people evaluate a video.
Source-stated
The authors say they release the study data to support further research on eye tracking for evaluating AI-generated video.
We release the study data to support further research on eye tracking for evaluating AI-generated video.
Source and contributor context
Shi, Danqing, Jiang, Lan, Lyons, Jake, Zilka, Miri · 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.
Machine Intelligence Research Institute
Explicit institutional argument for restricting superintelligence; useful strong opposing thesis.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: a4c69e68-4765-4bd2-8d08-72a380b05e05.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-11T09:08:11.790Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 6aa5b02618c0e3e2489d52db05d50282d0a711fed06729d03b34b34cfb44f1c4. Source text hash: 4f9c879b8cb2a3be67560c42b4b03325fd47144f7ec336bb3719379ebfc6a1cc.
Revision and source-check history
Version 1 · 2026-11-11T09:08:11.790Z
Researchers propose eye tracking to assess AI-generated video realism
Danqing Shi, Lan Jiang, Jake Lyons and Zilka argue that ratings of AI-generated videos capture viewers’ conclusions but not the perception leading to them. They study eye tracking as a complementary signal, recording where viewers look while watching videos. In two Visual Turing Test experiments, 90 participants watched videos generated by recent models twice and then gave feedback. The researchers report 4,286 valid eye-tracking trials across 438 videos, and say gaze patterns align with viewers’ explicit feedback. They introduce GazeEval, a computational method that predicts perceived visual realism from eye-tracking data. Its reported prediction accuracy is 71.3%; the authors present this as evidence that eye-tracking data carries information about how people evaluate a video. They release the study data for further research. The abstract does not establish how well the method would work with other viewers, videos or viewing conditions.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T09:08:11.790Z. First observed: 2026-10-30T06:43:01.472Z. Retrieved: 2026-10-31T06:32:57.905Z.
Limits
The available evidence is the paper’s abstract; it does not give details about participant characteristics, the videos or models tested, or how prediction accuracy was calculated. The abstract reports prediction accuracy and alignment with explicit feedback, but does not establish performance beyond the experiments described.
Open question
How well does GazeEval’s prediction of perceived visual realism hold beyond the 438 videos and 90 participants in these experiments?
Source-stated
The authors say explicit ratings capture viewers’ conclusions but not the perception behind them, and investigate eye tracking as a complementary signal.
While these signals capture the conclusions viewers reach, they fail to offer insights into the perception that leads to them.
Source-stated
Across two experiments, 90 participants watched videos twice and provided feedback; the researchers collected 4,286 valid eye-tracking trials on 438 videos.
Across the experiments, we collected 4,286 valid eye-tracking trials on 438 videos.
Source-stated
The authors report that gaze patterns provide informative evaluation signals that align with explicit human feedback.
Our analysis indicates that gaze patterns offer informative evaluation signals that align with explicit human feedback.
Source-stated
The authors propose GazeEval to predict perceived visual realism from eye-tracking data and report 71.3% prediction accuracy.
It achieves 71.3% accuracy in prediction, providing evidence that eye tracking data carries informative signals about how people evaluate a video.
Source-stated
The authors say they release the study data to support further research on eye tracking for evaluating AI-generated video.
We release the study data to support further research on eye tracking for evaluating AI-generated video.
Source and contributor context
Shi, Danqing, Jiang, Lan, Lyons, Jake, Zilka, Miri · 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.
Machine Intelligence Research Institute
Explicit institutional argument for restricting superintelligence; useful strong opposing thesis.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: a4c69e68-4765-4bd2-8d08-72a380b05e05.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-11T09:08:11.790Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 6aa5b02618c0e3e2489d52db05d50282d0a711fed06729d03b34b34cfb44f1c4. Source text hash: 4f9c879b8cb2a3be67560c42b4b03325fd47144f7ec336bb3719379ebfc6a1cc.
passed · 2026-11-11T09:08:11.790Z · anthropic/claude-opus-5.5
Publication: a4c69e68-4765-4bd2-8d08-72a380b05e05
attribution — pass: The reading correctly attributes the work to Danqing Shi, Lan Jiang, Jake Lyons and Miri Zilka and keeps the findings framed as the authors' reports and claims.
Authors:Danqing Shi, Lan Jiang, Jake Lyons, Miri Zilka
summary — pass: The title, summary and question match the abstract. Eye tracking is explained in plain language as recording where viewers look. Visual Turing Test is not defined, but the study design and its significance remain understandable without it. A term-library tooltip could be added. The generalization question follows from the stated scope.
This paper investigates eye tracking as a complementary human evaluation signal for AI-generated videos.
date — pass: The originalPublished value, 2026-10-28, matches the v1 submission date. The feed date of 2026-10-31 is later, which is expected.
[Submitted on 28 Oct 2026]
support — pass: Each claim is supported by a verbatim excerpt from the abstract. The 71.3% figure is presented as the authors' reported result, not as proof of real-world reliability. The limitations correctly note that the abstract does not establish generalization. No benefits or costs are invented.
It achieves 71.3% accuracy in prediction, providing evidence that eye tracking data carries informative signals about how people evaluate a video.
media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way for readers to access 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.
Explicit institutional argument for restricting superintelligence; useful strong opposing thesis.
Explore Machine Intelligence Research InstituteExplicit institutional argument for restricting superintelligence; useful strong opposing thesis.
Explore Machine Intelligence Research InstitutearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivExplicit institutional argument for restricting superintelligence; useful strong opposing thesis.
Explore Machine Intelligence Research InstitutearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv