“What would make artificial general intelligence general is generality: the ability to cope with new things. And that’s precisely where current AI is still lacking.”
Our reading, the sentence that was checked:
Generality, the ability to cope with new things, is what would make AGI general, and that is precisely where current AI is still lacking.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, No, Virginia, AGI is not imminent, paragraph 37.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-no-virginia-agi-is-not-imminent-15eaa7-a1
We recorded 17 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 2023-08-20. 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
- generality: the ability to cope with new things
- current AI
- still lacking
Our interpretation
- The checkable assertion is that contemporary AI had substantial limitations in handling unfamiliar tasks reliably. The opening statement supplies his meaning of generality.
- The plain reading allows successful generalization on some tasks. It does not assert that AI cannot solve anything new.
- Assess the latest publicly evaluated capabilities of systems available before the post on 2023-08-20. This is an as-of observation, not a reporting-period statistic; the source supplies no lower temporal bound, so target_period.start is unspecified.
- Use controlled tests of unfamiliar rules and conceptual variations as evidence of generalization. No single benchmark defines general intelligence.
Counts as supported if: Contemporary evaluations show substantial, systematic failures to transfer capabilities to unfamiliar task variants or to generalize concepts, including marked gaps against humans where comparable human results exist.
Counts as contradicted if: Broad contemporary evaluations show reliable adaptation across unfamiliar tasks and conceptual variations, making substantial generalization limitations an inaccurate description of leading AI systems.
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 contemporary evidence supports substantial limitations in generalization. ConceptARC found large human–machine gaps, and controlled changes to familiar task rules caused consistent performance drops across several language models. These findings directly address coping with unfamiliar situations. There was also successful transfer: better representations improved ARC performance, and exploratory GPT-4 work demonstrated broad capabilities. Those successes qualify the claim but do not overturn its ordinary reading that generalization remained unreliable. This assessment does not establish that AI lacked all generalization or that AGI could not arrive soon.
Limitations
- The outcome depends on reading 'lacking' as substantial limitations, rather than complete absence.
- Text representations and prompting affect measured performance.
- These are evaluations available before the post, not a comprehensive audit conducted on the publication day.
- The introductory definition of generality is treated as framing, not as a separately proven scientific definition.
Evidence (4 items)
Supports the claim: arXiv; Arseny Moskvichev, Victor Vikram Odouard and Melanie Mitchell,
40 percentage points.
Why it matters here: ConceptARC tested generalization across variations of concepts. Humans substantially exceeded specialized solvers; GPT-4 scored below 30% on 15 of 16 concepts, supporting a substantial limitation under these test conditions.
- Publisher
- arXiv; Arseny Moskvichev, Victor Vikram Odouard and Melanie Mitchell
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: arXiv; Zhaofeng Wu and colleagues,
consistently and substantially degrades
Why it matters here: Across 11 tasks, changing familiar rules exposed systematic transfer weaknesses in several contemporary models. Above-random performance on most changed tasks also showed that some generalization existed.
- Publisher
- arXiv; Zhaofeng Wu and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: arXiv; Yudong Xu and colleagues,
solves only 13/50
Why it matters here: The experiment demonstrates difficulty inferring transformations from limited examples. However, externally supplied object representations nearly doubled solved ARC tasks and produced near-perfect performance on easier one-dimensional tasks, showing that representation contributed materially to the failures.
- Publisher
- arXiv; Yudong Xu and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Sébastien Bubeck and colleagues, Microsoft Research,
human-level performance
Why it matters here: The authors reported broad successes on novel tasks in an early development version of GPT-4. This challenges an absolute inability reading, but the exploratory demonstrations do not establish reliable generalization across unfamiliar situations in the deployed systems.
- Publisher
- arXiv; Sébastien Bubeck and colleagues, Microsoft Research
- 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, No, Virginia, AGI is not imminent
What would make artificial general intelligence general is generality: the ability to cope with new things. And that’s precisely where current AI is still lacking.
- Where
- paragraph 37
- Published
- 2023-08-20
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-no-virginia-agi-is-not-imminent-15eaa7-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-no-virginia-agi-is-not-imminent-15eaa7-a1.