“Scale isn’t everything, and hasn’t and won’t be enough to solve any of the problems I just listed above.”
Our reading, the sentence that was checked:
Scaling will not be enough to solve the problems of understanding, compositionality, negation, and reliability in LLMs.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, There Must be Some Misunderstanding, paragraph 10.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-there-must-be-some-misunderstanding-8e10c2-2-a1
We recorded 7 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 2024-02-08. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
From the speaker
- Scale isn’t everything
- won’t be enough
- to solve any of the problems I just listed above
Our interpretation
- Assess the future 'won't' component separately from the historical observation.
- Scale means increasing model size, training data and compute within substantially the same method. Material architecture or training-method changes cannot automatically be credited to scale alone.
- Solving a problem means dependable performance across varied, relevant tasks and novel cases, not success on one example or a narrow benchmark.
- Because the claim says 'any', a convincing scaling-only resolution of even one listed problem could contradict it.
- No deadline is supplied. Continuing failures cannot establish that scaling will never suffice.
Counts as supported if: A general result establishing that scaling alone cannot resolve any of the listed problems under assumptions that actually cover the claimed model class.
Counts as contradicted if: A publicly documented, independently reproduced model or model family broadly resolves at least one listed problem, with controlled evidence attributing that resolution to scaling alone.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
No reviewed result meets the early-resolution rule. Scaling has produced gains, including compositional generalization on controlled tasks, while other tests retain substantial gaps. Neither finding settles whether scaling alone can eventually solve a listed problem broadly. With no deadline, continued failures cannot establish the forecast.
Limitations
- The boundary between scaling and a new method is unclear.
- A narrow benchmark reading would give greater weight to the positive compositionality result.
- The search cannot establish the absence of every potentially relevant result.
Evidence (3 items)
Context only: arXiv; Apoorv Khandelwal and Ellie Pavlick,
the gap clearly remains
Why it matters here: The study reports smaller compositionality gaps at greater model sizes, but substantial remaining gaps. These finite results establish neither permanent impossibility nor broad resolution through scaling.
- Where
- Section 3.2, Results
- Publisher
- arXiv; Apoorv Khandelwal and Ellie Pavlick
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Simon Schug and colleagues,
Scaling model size and data enables compositional generalization.
Why it matters here: Controlled symbolic reasoning experiments show meaningful compositional generalization from scaling, with 80% success on held-out task instances at sufficient scale. This challenges a blanket denial of scaling benefits, but does not establish broad resolution of compositionality in LLMs.
- Where
- Section 4.4, Results
- Publisher
- arXiv; Simon Schug and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
along with architecture and optimization innovations.
Why it matters here: OpenAI reports improved knowledge and reliability from a larger model, while explicitly identifying additional innovations. This announcement establishes neither elimination of a listed problem nor attribution to scaling alone.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
An independently reproduced result showing that scaling alone broadly resolves at least one listed problem would contradict the forecast. An applicable general impossibility result covering all listed problems could support it; further finite failures would leave it open.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, There Must be Some Misunderstanding
Scale isn’t everything, and hasn’t and won’t be enough to solve any of the problems I just listed above.
- Where
- paragraph 10
- Published
- 2024-02-08
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-there-must-be-some-misunderstanding-8e10c2-2-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-there-must-be-some-misunderstanding-8e10c2-2-a1.