“There’s always a better model. o3 does better on some of the examples in my original article, but still can’t do my new (hence not previously discussed) example of the woman and the watch:”
Our reading:
o3 does better on some old examples but still cannot correctly generate the new woman-and-watch image.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
The claim concerns particular reported tests. The retrieved post supplies Marcus’s account, but the permitted evidence does not independently verify the outputs or the comparison with his earlier examples. Other users’ generations cannot establish or disprove what happened in his session.
Caveats
- The page reader did not expose the images needed for visual assessment.
- Missing corroboration is not evidence that the reported failure did not occur.
Evidence (0 items)
Nothing we found settles this yet.
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Conditions
- new, not previously discussed example (their words)
Where we had to interpret
- This describes Marcus’s reported tests, not an assertion that every possible generation must fail.
- Improvement means better adherence to some prompts from his original article compared with the earlier outputs.
- The woman-and-watch output must jointly depict left-handed writing, a watch on the right wrist, the time 3:20, and the requested inscription.
- New means not previously discussed in his published examples; it does not establish absence from training data.
- The target is the testing reported by June 25, 2025. The exact testing dates are unspecified, so the start is left unknown.
Right if: Authenticated records of the reported tests show improved adherence on some original prompts and failure to satisfy all requirements in the reported woman-and-watch attempt or attempts.
Wrong if: Authenticated records of those same tests show that the alleged failed woman-and-watch output satisfied all requirements, or establish that the claimed improvements on the original examples did not occur. A successful generation in another session cannot contradict the reported failure.
We wrote this test after the outcome was known.
What would settle it
An authenticated record of the reported prompts and outputs, with model identification and the earlier comparison images, independently inspected against each requested detail.
We found 8 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Image generation: Still crazy after all these years, paragraph 30.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2025-06-25.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Image generation: Still crazy after all these years
There’s always a better model. o3 does better on some of the examples in my original article, but still can’t do my new (hence not previously discussed) example of the woman and the watch:
- Where
- paragraph 30
- Published
- 2025-06-25
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
A reviewer said our test wasn’t fair to what was actually said.