“But one important component of physical and conceptual understanding is reasoning about entities and their properties, and current approaches have consistently fallen short on this.”
Our reading, the sentence that was checked:
Current AI approaches have consistently fallen short on reasoning about entities and their properties.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI still lacks “common” sense, 70 years later, paragraph 43.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-ai-still-lacks-common-sense-70-years-a13f79-a1
We recorded 15 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 2025-01-05. 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)
From the speaker
- physical and conceptual understanding
- reasoning about entities and their properties
- current approaches
- have consistently fallen short
Our interpretation
- The ordinary reading is recurring shortcomings in dependable reasoning about objects, people, properties, and changes, rather than failure on every example.
- Current approaches refers principally to the language-based and video-based systems discussed in the post.
- Use the latest completed model-generation reporting year before publication, January 2024 through January 5, 2025, for this ongoing capability assessment.
- The proposition that entity reasoning is important supplies context; the scored empirical assertion concerns the approaches' shortcomings.
Counts as supported if: Contemporary primary evaluations document recurring errors in entity tracking or property changes across language-based and video-based approaches.
Counts as contradicted if: Representative evaluations show dependable entity and property reasoning across those approaches, making recurring shortcomings an inaccurate description.
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 evidence supports recurring shortcomings rather than universal failure. Sora's developer acknowledged incorrect object-state changes, while a multi-family language-model evaluation found substantial errors even in its strongest entity tracker. Both sources also document real capabilities or improvements. Those successes qualify the claim's breadth but do not establish dependable reasoning across the situations tested.
Limitations
- This outcome depends on reading consistently as recurring gaps, not failure on every task.
- The language-model study does not cover all frontier models available at the post date.
- Sora's disclosure supplies examples and qualitative findings rather than a comprehensive error rate.
Evidence (2 items)
Supports the claim: OpenAI,
correct changes in object state.
Why it matters here: OpenAI directly acknowledges incorrect object-state changes in Sora. The report also describes successful object persistence and some physical interactions, qualifying any reading that denies all entity-related competence.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: arXiv / Kim, Schuster, and Toshniwal,
64.9 accuracy on 1+ operations
Why it matters here: The best tested model still made substantial errors on nontrivial entity tracking. Comparisons across model families also show that code training improves performance, which is meaningful counterevidence to a stronger claim of no progress or no competence.
- Publisher
- arXiv / Kim, Schuster, and Toshniwal
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, AI still lacks “common” sense, 70 years later
But one important component of physical and conceptual understanding is reasoning about entities and their properties, and current approaches have consistently fallen short on this.
- Where
- paragraph 43
- Published
- 2025-01-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ai-still-lacks-common-sense-70-years-a13f79-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-ai-still-lacks-common-sense-70-years-a13f79-a1.