“Text in general is better now (though not in this image) but watch comprehension of any time other than 10:10 (most frequent in advertisements) remains dreadful.”
Our reading, the sentence that was checked:
Imagen 4 is better at text in general now, but its handling of watch times other than 10:10 remains dreadful.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Image generation: Still crazy after all these years, paragraph 25.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-image-generation-still-crazy-after-20666b-a2
We recorded 4 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 2025-06-25. 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)
Conditions
- times other than 10:10 (qualifier, stated by the speaker)
- text better in general, though not in this image (qualifier, stated by the speaker)
From the speaker
- Text in general is better now
- though not in this image
- watch comprehension of any time other than 10:10
- remains dreadful
Our interpretation
- Text improvement refers to earlier Imagen models, with original Imagen providing the surrounding historical baseline.
- Watch comprehension means generating an analog watch whose hands display the requested time.
- Dreadful ordinarily means highly unreliable performance across requested times other than 10:10, not mathematical impossibility of a single correct output.
- A representative evaluation should include 3:20 and several other non-10:10 times, with repeated unselected generations.
- The target is Imagen 4 as used in the June 24–25 testing episode; no later product version is substituted.
Counts as supported if: Comparable evidence establishes improved text rendering, while representative tests of the relevant Imagen 4 version show that most requested non-10:10 watch times are incorrect.
Counts as contradicted if: Representative tests establish reliable rendering across non-10:10 watch times, or comparable evaluations refute the text improvement. Opposing established findings across the two components would warrant mixed.
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
The clock study generated images on October 7, 2026, and reports results across multiple models (e3). It does not establish Imagen 4’s watch-time accuracy or improvement in text rendering during June 24–25, 2025, so neither component is settled (e3).
Limitations
- The packet contains no representative watch-time evaluation matched to the June 2025 Imagen 4 version.
- The packet contains no comparable evaluation of text rendering against earlier Imagen models.
- The later study’s multi-clock task does not establish performance on the requested watch task.
Evidence (1 item)
Context only: arXiv,
2:35, 6:50 and 11:20.
Why it matters here: This is a relevant test design for non-10:10 times, published before the evaluation cutoff. Its generation date is outside the target period, and its abstract does not establish the June 2025 Imagen 4 watch accuracy.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Archived, independently scored tests of the June 2025 Imagen 4 version, with recorded prompts and all outputs, covering repeated generations of 3:20 and several other non-10:10 watch times, plus comparable text-rendering tests against earlier Imagen models.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Image generation: Still crazy after all these years
Text in general is better now (though not in this image) but watch comprehension of any time other than 10:10 (most frequent in advertisements) remains dreadful.
- Where
- paragraph 25
- Published
- 2025-06-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-image-generation-still-crazy-after-20666b-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-image-generation-still-crazy-after-20666b-a2.