“Video-generators (even the latest ones) too are insufficiently anchored in the world.”
Our reading, the sentence that was checked:
Even the latest video generators, such as Veo 3, are not anchored firmly enough in the world.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Generative AI’s crippling and widespread failure to induce robust models of the world, paragraph 64.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-generative-ais-crippling-and-widespread-1a07e1-a1
We recorded 18 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-06-28. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- even the latest ones (qualifier, stated by the speaker)
From the speaker
- Video-generators
- even the latest ones
- insufficiently anchored in the world
- Veo 3
Our interpretation
- The plain behavioral reading is that contemporary leading generators, including Veo 3, still produce recurring physical or entity-consistency errors in scenes intended to depict coherent events.
- This does not independently establish whether the systems possess internal world representations.
- Because no reporting period is specified, use the current calendar quarter through publication, April 1–June 28, 2025.
- Deliberately requested fantasy or impossible scenes do not count as failures merely because they violate real-world physics.
Counts as supported if: Systematic evaluations show recurring physical-consistency failures across contemporary generators, with direct evidence that the limitation also affects Veo 3.
Counts as contradicted if: Broad tests of contemporary leading systems, including Veo 3, show reliable physical and entity consistency across challenging realistic scenes, with alleged failures explained by prompts requesting impossible content or invalid testing.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The behavioral reading is supported. A contemporary benchmark documents recurring physical failures across generators, and firsthand Veo 3 tests show that the new model also produces implausible body and scene changes. Google's account of stronger realism is compatible with these remaining failures. The evidence does not establish Marcus's broader theory about internal world models.
Limitations
- The result depends on interpreting anchored as observable physical and entity consistency.
- Veo 3 evidence is a documented hands-on test, not a representative benchmark.
- The benchmark's May reporting date does not establish exact generation dates.
- Successful realistic outputs and substantial improvements coexist with the documented errors.
Evidence (3 items)
Supports the claim: arXiv; Xuyang Guo and colleagues,
no system achieves even moderate accuracy
Why it matters here: T2VPhysBench reports low physical-compliance scores across ten generators using human ratings and prompts testing physical laws. Its ordinary-prompt evaluation supports the class-level observation, but it predates Veo 3 and cannot establish that model's performance.
- Publisher
- arXiv; Xuyang Guo and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Ars Technica,
occasional glitches
Why it matters here: The reporter's own tests include a realistic gymnastics prompt producing implausible changes to body parts and problems maintaining scene entities. This is firsthand evidence about Veo 3; it is used because the selected research benchmark predates that model, although it does not estimate a representative failure rate.
- Publisher
- Ars Technica
- Kind of source
- News report
- Describes events of
- Public from
Context only: Google,
real-world physics
Why it matters here: Google reports substantial capability in physical realism and prompt following. This counters a reading that Veo 3 cannot depict coherent scenes, but does not establish reliable consistency on difficult prompts.
- Publisher
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Generative AI’s crippling and widespread failure to induce robust models of the world
Video-generators (even the latest ones) too are insufficiently anchored in the world.
- Where
- paragraph 64
- Published
- 2025-06-28
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-generative-ais-crippling-and-widespread-1a07e1-a1: Current. Outcome: Supported. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-generative-ais-crippling-and-widespread-1a07e1-a1.