“Negation is, as ever, a problem.”
Our reading, the sentence that was checked:
Text-to-image systems like Stable Diffusion continue to have problems with negation in prompts.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Form, function, and the giant gulf between drawing a picture and understanding the world, paragraph 34.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-form-function-and-the-giant-gulf-df5e53-a1
We recorded 6 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2022-09-18. 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
- Negation is, as ever, a problem.
- Draw a white bicycle with no wheels
- Stable Diffusion
Our interpretation
- The ordinary reading is that contemporary Stable Diffusion repeatedly mishandles negation in ordinary text prompts. It does not assert that every negated prompt fails.
- The example concerns negation within the main prompt, rather than a separate negative-prompt control.
- The post describes ongoing September 2022 behavior without a reporting period. I use September through the publication date as the observation window.
- Later experiments are context unless their records establish behavior within that observation window.
Counts as supported if: Independent records establish recurring failures by contemporary Stable Diffusion to obey ordinary negated prompts during the target period, including comparable object-exclusion requests.
Counts as contradicted if: Representative independent tests during the target period establish reliable compliance with ordinary negated prompts, including the bicycle example and comparable exclusions.
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 independent study is consistent with Marcus's observation, but its experiments are not dated to the September observation window. Under the required event-date rule, it supplies context rather than decisive outcome evidence. The searches did not locate qualifying contemporary tests that establish or refute recurring negation failures.
Limitations
- This is an evidence gap, not a finding that negation worked reliably.
- The inferred observation window materially limits which experiments qualify.
- Later tests of the same checkpoint could support a broader retrospective assessment, but their historical applicability has not been established here.
Evidence (1 item)
Context only: arXiv; Shashank Sonkar, Naiming Liu, and Richard G. Baraniuk,
function words are poorly understood
Why it matters here: The paper tests Stable Diffusion v1.4 and reports failures with negated descriptions. It does not give experiment dates establishing that these observations occurred within September 1–18, so it is contextual evidence only.
- Publisher
- arXiv; Shashank Sonkar, Naiming Liu, and Richard G. Baraniuk
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Independent dated outputs or experimental records from September 1–18, 2022, identifying the Stable Diffusion checkpoint and showing repeated results for ordinary negated prompts.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Form, function, and the giant gulf between drawing a picture and understanding the world
Negation is, as ever, a problem.
- Where
- paragraph 34
- Published
- 2022-09-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-form-function-and-the-giant-gulf-df5e53-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-form-function-and-the-giant-gulf-df5e53-a1.