“As we can see, multimodal training is not magically making hallucinations go away.”
Our reading, the sentence that was checked:
Multimodal training is not making hallucinations go away.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Hello, Multimodal Hallucinations, paragraph 11.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-hello-multimodal-hallucinations-d71f3a-a2
We recorded 11 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 2023-10-21. 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
- multimodal training
- not magically making hallucinations go away
- As we can see
Our interpretation
- The plain reading is that adding visual training had not eliminated hallucinations in contemporary multimodal models.
- The claim does not say that multimodal training offers no improvement or that elimination is forever impossible.
- I assess persistence after multimodal training, not its isolated causal effect relative to otherwise identical text-only training.
- For the ongoing state of published evaluations, I use January 1–October 21, 2023, the current research reporting year through publication.
Counts as supported if: Contemporaneous multimodal models exhibit verified hallucinations after training on visual and language inputs.
Counts as contradicted if: Representative contemporary evaluations establish that hallucinations had been eliminated in the relevant multimodal systems, with the alleged failures explained as something else.
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
OpenAI’s September 25, 2023 GPT-4V system card documented hallucinations during developer testing [e1]. An independent preprint published May 17, 2023 reported objects inconsistent with the supplied images across several vision-language models [e2]. These findings support the observation that multimodal training had not eliminated hallucinations in contemporary models by October 21, 2023 [e1, e2].
Limitations
- The evidence does not isolate the causal effect of multimodal training or establish that it cannot reduce hallucination rates.
- The findings concern contemporary models in 2023 and do not establish that future elimination is impossible.
- The packet does not independently verify the Julia Child kitchen exchange.
Evidence (2 items)
Supports the claim: OpenAI,
prone to hallucinations
Why it matters here: The release assessment reports hallucinations in GPT-4V itself. Its Be My Eyes discussion also reports reductions in hallucination frequency and severity during testing, supporting improvement without elimination.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: arXiv; Yifan Li and colleagues,
hallucination
Why it matters here: Independent experiments found object hallucinations across several vision-language models. This supports persistence beyond the single kitchen anecdote.
- Publisher
- arXiv; Yifan Li and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Hello, Multimodal Hallucinations
As we can see, multimodal training is not magically making hallucinations go away.
- Where
- paragraph 11
- Published
- 2023-10-21
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-hello-multimodal-hallucinations-d71f3a-a2: 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-hello-multimodal-hallucinations-d71f3a-a2.