“And guess what, surprise, surprise, even the latest models still struggle with compositionality”
Our reading, the sentence that was checked:
Even the latest image generation models, Nano Banana among them, still struggle with compositionality.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What scaling does and doesn’t buy you: peeling back the hype surrounding Google‘s trendy Nano Banana, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-scaling-does-and-doesnt-buy-6b861f-a1
We recorded 8 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 2025-08-27. 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
- even the latest models
- still struggle with compositionality
- nano
Our interpretation
- Compositionality means correctly combining requested objects, attributes, counts, spatial relations, and parts.
- The ordinary reading is that meaningful compositional difficulties persist in frontier image generators, including Nano Banana. It does not imply universal failure or no improvement.
- For this ongoing observation, use August 2025 through publication, covering contemporary public tests of the models under discussion.
Counts as supported if: Contemporaneous evaluations document recurring substantive compositional errors in Nano Banana and other frontier image generators on clearly specified prompts.
Counts as contradicted if: Representative contemporaneous evaluations show reliable compositional performance across the relevant models, with the alleged difficulties attributable to ambiguous prompts, misidentified models, or unrepresentative examples.
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
A launch-day independent test reports compositional failures in Nano Banana, alongside successes. Google also acknowledges remaining factual-detail limitations. These records provide some support, but do not establish recurring difficulties across the latest models under the contract. The broader claim remains unresolved; this is not evidence that compositionality was solved.
Limitations
- The independent test is small and selected, and its outputs were not independently rescored.
- Company statements about factual detail are not compositionality measurements.
- Successful prompts in e1 limit any reading that Nano Banana generally cannot compose unfamiliar combinations.
Evidence (2 items)
Supports the claim: Apolinário (@multimodalart),
nano banana
Why it matters here: The accompanying results mark a brim-full wine glass and a seven-fingered hand as failures, while marking a horse riding an astronaut and a blue-yolked egg as successes. This supports persistent difficulties in Nano Banana, but supplies neither repeated trials nor a contemporary multi-model comparison.
- Publisher
- Apolinário (@multimodalart)
- Kind of source
- Demonstration
- Describes events of
- Public from
Context only: Google Developers Blog,
factual
Why it matters here: Google acknowledges remaining limitations at launch. Factual detail and character consistency overlap with parts of the claim, but this disclosure does not measure compositional performance.
- Publisher
- Google Developers Blog
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Accessible dated tests from the target period covering Nano Banana and other contemporary frontier image generators, with identifiable versions, prompts, outputs, and enough trials to distinguish recurring difficulties from isolated failures.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What scaling does and doesn’t buy you: peeling back the hype surrounding Google‘s trendy Nano Banana
And guess what, surprise, surprise, even the latest models still struggle with compositionality
- Where
- paragraph 4
- Published
- 2025-08-27
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-scaling-does-and-doesnt-buy-6b861f-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-what-scaling-does-and-doesnt-buy-6b861f-a1.