“Hoping that reliable commonsense will simply “emerge” from large databases and video has proven unrealistic, again and again.”
Our reading, the sentence that was checked:
The hope that reliable commonsense will emerge from large text and video datasets has repeatedly proven unrealistic.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AI still lacks “common” sense, 70 years later, paragraph 42.
- 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-f53ad1-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
- reliable commonsense
- simply “emerge”
- large databases and video
- has proven unrealistic
- again and again
Our interpretation
- The ordinary reading is that large-scale text and video learning had repeatedly fallen short of dependable commonsense across varied situations by publication.
- The claim does not deny useful learned knowledge or success on particular tasks, and it does not establish permanent impossibility.
- Repeated evidence must cover multiple evaluations and both language-based and video-based approaches.
- The retrospective period runs from 2020, when the surrounding discussion introduces GPT-3, through January 5, 2025.
Counts as supported if: Multiple evaluations within the period document recurring commonsense failures in language-based systems and failures of physical generalization in video-based systems.
Counts as contradicted if: Representative evaluations establish broad, repeatable commonsense reliability in both model classes, making the claimed pattern of shortcomings inaccurate.
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 historical reading is supported. Separate evaluations document problems with changing entity states, implicit workplace assumptions, and physical predictions beyond familiar video examples. These are repeated shortcomings across the approaches discussed. The video study also reports strong successes in familiar settings, so the evidence supports unreliability across situations, not an absence of learned capabilities or a proof that scaling can never work.
Limitations
- Reliable commonsense remains an interpretive standard.
- The video experiments use simplified physical environments rather than every commercial video model.
- The studies are samples of capabilities, not an exhaustive evaluation of all AI.
- The January 2024 item dates the reported evaluation version; individual test execution dates are not specified.
Evidence (3 items)
Supports the claim: arXiv / Zhang and colleagues,
far from competent.
Why it matters here: The reported evaluation concerns entity and attribute changes in procedural text, a concrete component of everyday reasoning. It establishes a shortcoming in that task family rather than a universal absence of commonsense.
- Publisher
- arXiv / Zhang and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: arXiv / Xu and colleagues,
infer implicit assumptions.
Why it matters here: The study identifies commonsense-related failures in a multi-model workplace-agent evaluation. Its overall task failures have several causes, so they cannot all be counted as commonsense failures.
- Publisher
- arXiv / Xu and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: arXiv / Kang and colleagues,
failure in out-of-distribution scenarios.
Why it matters here: Controlled video experiments found that scaling did not deliver dependable physical predictions outside the training distribution. The same results show substantial successes within familiar distributions and some improvements on new combinations.
- Publisher
- arXiv / Kang and colleagues
- 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
Hoping that reliable commonsense will simply “emerge” from large databases and video has proven unrealistic, again and again.
- Where
- paragraph 42
- Published
- 2025-01-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-ai-still-lacks-common-sense-70-years-f53ad1-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-f53ad1-a1.