“What’s missing is any global comprehension by the system.”
Our reading, the sentence that was checked:
In most or all of the Sora errors seen so far, virtually every bit of the images is locally correct, and what's missing is any global comprehension by the system.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Sora Can’t Handle the Truth, paragraph 16.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-sora-cant-handle-the-truth-1ae935-a1
We recorded 11 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2024-02-17. 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
- in most or all of the errors seen thus far (scope, stated by the speaker)
- virtually every bit (qualifier, stated by the speaker)
From the speaker
- any global comprehension by the system
- most or all of the errors we have seen thus far
- virtually every bit of the images are locally correct
- a tiny patch
- for a single moment
Our interpretation
- The ordinary contextual reading of global comprehension is representing and maintaining scene-wide relationships, object identities, and counts across space and time.
- This concerns the original Sora demonstrations available by the post date, not later versions.
- The surrounding local-correctness claim concerns most observed faulty clips, rather than all possible Sora outputs.
- The absence of any such comprehension is stronger than occasional failures of scene consistency. Neither an isolated failure nor an isolated convincing output settles the internal-representation claim.
Counts as supported if: A defined collection of the contemporaneous faulty demonstrations establishes predominantly correct local appearance but scene-wide failures, and model evidence supports the claimed absence of representations maintaining those relationships.
Counts as contradicted if: Evidence directly establishes relevant scene-wide representations in the original model, or a representative examination refutes the claimed predominance of locally correct content in its faulty demonstrations.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
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
Scene-consistency failures do not by themselves establish that the system lacks any scene-wide representation. OpenAI also reports spatial and temporal consistency capabilities, which complicate that inference without proving comprehension. The retrieved record supplies neither a systematic local-versus-global error assessment nor a decisive study of the original model’s representations. The stronger claim therefore remains unsettled.
Limitations
- The outcome depends on preserving the strength of any global comprehension rather than reducing it to occasional inconsistency.
- Company demonstrations are selected examples, not a representative evaluation.
- Later models cannot settle the internal properties of the February 2024 model.
Evidence (1 item)
Context only: OpenAI,
dynamic camera motion.
Why it matters here: The section describes coherent movement through space; the following section reports some persistence through occlusion and across shots. These reported successes challenge an inference of complete absence from failures, but do not directly establish internal comprehension.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A defined assessment of the original demonstrations and a controlled study of the original model’s scene-wide representations, using explicit criteria for local correctness and global consistency.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Sora Can’t Handle the Truth
What’s missing is any global comprehension by the system.
- Where
- paragraph 16
- Published
- 2024-02-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-sora-cant-handle-the-truth-1ae935-a1: Current. Outcome: Unresolved. 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-sora-cant-handle-the-truth-1ae935-a1.