“Compositionality remains a problem. So does negation.”
Our reading, the sentence that was checked:
Compositionality and negation remain problems for large language models.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, There Must be Some Misunderstanding, paragraph 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-there-must-be-some-misunderstanding-8ebd86-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 2024-02-08. 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
- Compositionality remains a problem.
- So does negation.
- large language models
Our interpretation
- The ordinary reading is that contemporary LLMs still showed systematic failures, not that they invariably failed either capability.
- Compositionality means correctly combining component operations or instructions. Negation includes following negatively phrased instructions.
- For this ongoing observation, use the latest completed calendar year and the publication-year period through the post as the reporting window. These capabilities have no standard quarterly reporting cycle.
- The surrounding examples concern generated images, but this assessment follows the supplied proposition about LLM capabilities. It does not independently establish image-generation failure rates.
Counts as supported if: Contemporary model evaluations document repeatable, material failures in both compositional tasks and negation.
Counts as contradicted if: Broad, reproducible evaluations show that both capabilities were dependable across varied tasks and ordinary prompts, with the alleged failures explained by defective tests or obsolete models.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
Contemporary evaluations found systematic compositional failures and a clear negative-instruction failure in GPT-4. This supports the modest statement that both remained problems. The negation evidence also shows substantial successes, so it does not support a stronger claim that LLMs generally cannot understand negation.
Limitations
- Results depend on task, model version and prompt.
- The negation experiment is an original public test, not a peer-reviewed study.
- Persistence at publication is inferred from the contemporary evaluation record.
Evidence (2 items)
Supports the claim: Allen Institute for Artificial Intelligence,
a significant decline in performance from nearly perfect to zero
Why it matters here: The researchers tested multiplication, logic puzzles and dynamic programming. GPT-4's multiplication accuracy declined sharply with problem complexity; step-by-step prompting improved simpler cases without eliminating the harder failures.
- Publisher
- Allen Institute for Artificial Intelligence
- Kind of source
- Independent measurement
- Describes events of
- Public from
Supports the claim: Alex Bleakley,
GPT-4 scores 37% with positive phrasing, but 0% with the negative instruction
Why it matters here: Paired tests document a repeatable negative-instruction failure in GPT-4. However, other tasks succeeded, and GPT-3.5 and Claude 2.0 performed similarly on positive and negative instructions overall; this limits any blanket claim of inability.
- Publisher
- Alex Bleakley
- Kind of source
- Independent measurement
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, There Must be Some Misunderstanding
Compositionality remains a problem. So does negation.
- Where
- paragraph 7
- Published
- 2024-02-08
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-there-must-be-some-misunderstanding-8ebd86-a1: Current. Outcome: Supported. 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-there-must-be-some-misunderstanding-8ebd86-a1.