“It’s the same issue of canonical versus noncanonical statistics, all over again.”
Our reading, the sentence that was checked:
Multimodal ChatGPT's trouble telling time from watch images comes from the same canonical-versus-noncanonical statistics problem seen in DALL-E.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Race, statistics, and the persistent cognitive limitations of DALL-E, paragraph 10.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-race-statistics-and-the-persistent-c2f239-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 2023-10-14. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- the same issue
- canonical versus noncanonical statistics
- all over again
Our interpretation
- Read with the surrounding watch discussion: common advertised watch configurations are claimed to explain multimodal ChatGPT's erroneous readings.
- Assess the October 2023 multimodal ChatGPT system, not later models merely sharing the ChatGPT name.
- A tendency to answer a common time is evidence consistent with the explanation, but does not by itself establish the training-distribution cause.
- Use October 1–14, 2023 as the contemporaneous window; the post describes a discovery a couple of days earlier without supplying an exact experiment date.
Counts as supported if: Controlled contemporaneous watch-reading tests establish a systematic canonical-time error pattern, and training-distribution evidence or interventions substantiate that pattern as the explanation for the failures.
Counts as contradicted if: Controlled tests of the relevant system establish that canonical-time statistics do not explain the errors, for example by isolating a different mechanism while varying canonical-time prevalence.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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
Later controlled work documents clock-reading problems and possible perceptual contributors. It does not establish why October 2023 multimodal ChatGPT failed. A common-time response pattern would be consistent with Marcus's account, but the located evidence does not connect that pattern to the relevant training distribution through a causal test.
Limitations
- Later model evaluations cannot be substituted for the historical deployment.
- The quoted study suggests a contributor rather than establishing an exclusive alternative cause.
- The original demonstration's experimental details were not recovered.
Evidence (1 item)
Context only: Saxena, Gema and Minervini; arXiv,
hand detection and angle interpretation.
Why it matters here: The later study reports clock-reading failures and identifies hand detection and angle interpretation as possible contributors. This cautions against inferring a unique training-frequency cause, but does not disprove Marcus's explanation for the earlier system.
- Publisher
- Saxena, Gema and Minervini; arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Archived controlled tests of the October 2023 system, combined with training-distribution measurements or interventions that isolate canonical-time frequency from visual-perception difficulty.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Race, statistics, and the persistent cognitive limitations of DALL-E
It’s the same issue of canonical versus noncanonical statistics, all over again.
- Where
- paragraph 10
- Published
- 2023-10-14
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-race-statistics-and-the-persistent-c2f239-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-race-statistics-and-the-persistent-c2f239-a1.