“It took me less than an hour to find half a dozen clear counterexamples in which ChatGPT 4o failed with respect to compositionality, evident even to a teenager.”
Our reading, the sentence that was checked:
In less than an hour, Marcus found half a dozen clear counterexamples in which ChatGPT 4o failed with respect to compositionality, failures evident even to a teenager.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Scott Alexander’s Misleading Victory Lap, paragraph 39.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-scott-alexanders-misleading-victory-303f61-a2
We recorded 6 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 2025-07-09. 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
- It took me less than an hour
- half a dozen clear counterexamples
- ChatGPT 4o failed
- with respect to compositionality
- evident even to a teenager
Our interpretation
- Half a dozen means six qualifying counterexamples found within less than 60 minutes.
- A counterexample is a prompt-output pair with a clear failure to satisfy a compositional instruction; multiple errors in one image do not automatically count as multiple examples.
- Evident even to a teenager ordinarily means recognizable without specialist knowledge, not that an actual teenage panel participated.
- The claim concerns Marcus's particular discovery session before publication. Its starting date is unspecified and is left unknown.
Counts as supported if: Independently verifiable records establish that Marcus found six distinct qualifying prompt-output failures within a discovery interval shorter than 60 minutes, with errors clear to non-specialist readers.
Counts as contradicted if: Complete records establish that the referenced discovery interval lasted at least 60 minutes, contained fewer than six qualifying examples, or relied on outputs that did not clearly violate their prompts.
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
Davis's July 2025 report documents a bicycle-chain error in his own GPT-4o test [e1]. That test cannot establish or refute whether Marcus found six qualifying failures in under an hour before publication on July 9, 2025 [e1; marcus-scott-alexanders-misleading-victory-303f61-p1]. The packet contains no independently verifiable record of Marcus's session, leaving his report unresolved [marcus-scott-alexanders-misleading-victory-303f61-p1].
Limitations
- The evidence does not establish Marcus's discovery interval, identify all six claimed prompt-output pairs, or verify their model identity and clarity to non-specialists.
- Davis's experiment cannot verify Marcus's count or discovery time.
- Missing session records do not establish that Marcus's report was false.
Evidence (1 item)
Context only: Ernest Davis, New York University,
The chain still isn't right.
Why it matters here: Davis documents an unresolved bicycle-chain error in his own GPT-4o experiment. This is context for the existence of related failures, not evidence of Marcus finding six examples within an hour.
- Publisher
- Ernest Davis, New York University
- Kind of source
- Demonstration
- Describes events of
- Public from
What would settle it
An authenticated, complete record of Marcus's discovery session showing timestamps, model identity, original prompts and relevant outputs, sufficient to establish whether he found six distinct compositional failures clear to non-specialists within less than 60 minutes.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Scott Alexander’s Misleading Victory Lap
It took me less than an hour to find half a dozen clear counterexamples in which ChatGPT 4o failed with respect to compositionality, evident even to a teenager.
- Where
- paragraph 39
- Published
- 2025-07-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-scott-alexanders-misleading-victory-303f61-a2: 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-scott-alexanders-misleading-victory-303f61-a2.