“On overt measures, the systems were fine. On covert measures, they were a disaster:”
Our reading, the sentence that was checked:
LLMs tested looked fine on overt racism measures but did very badly on covert racism measures when prompted with African American English.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Covert racism in LLMs, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-covert-racism-in-llms-b36446-a2
We recorded 8 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2024-03-05. 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)
Conditions
- given African American English prompts (qualifier, stated by the speaker)
- covert rather than overt measures (qualifier, stated by the speaker)
From the speaker
- On overt measures
- the systems were fine
- On covert measures, they were a disaster
Our interpretation
- The surrounding discussion identifies the systems and measures as those in Hofmann and colleagues’ newly released study.
- The ordinary reading is a marked contrast between generally favorable overt associations and strongly negative dialect-based associations. 'Fine' is not interpreted as a guarantee of zero overt bias.
- The target is the study reported before the post. Its experimental dates are unspecified, so the target period has no invented lower bound.
Counts as supported if: The cited study finds generally favorable overt associations with African Americans alongside strongly negative associations with speakers of African American English.
Counts as contradicted if: The study finds no substantial covert prejudice, or finds overt associations comparably negative rather than the claimed contrast.
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
The Nature abstract, published August 28, 2024, reports negative dialect-based stereotypes about speakers of African American English and describes overt stereotypes about African Americans as “more positive” [e1]. This supports the claimed contrast in the study’s measures, with “fine” understood as comparatively favorable rather than free of overt bias [e1].
Limitations
- The abstract does not establish that every tested model was free of overt bias.
- The findings do not establish how all LLMs behave in unrestricted conversations or actual consequential decisions.
- The cited abstract appeared after Marcus’s March 5 post; e1 identifies it as corroborating the findings originally reported March 1, rather than reporting a new August experiment.
Evidence (1 item)
Supports the claim: Nature,
the language models’ overt stereotypes about African Americans are more positive.
Why it matters here: The abstract contrasts strongly negative dialect-based stereotypes with more positive overt stereotypes. This retrospectively corroborates the findings already reported in the March 1 preprint; the event date identifies that original reporting, not a new August experiment.
- Where
- Abstract
- Publisher
- Nature
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Covert racism in LLMs
On overt measures, the systems were fine. On covert measures, they were a disaster:
- Where
- paragraph 4
- Published
- 2024-03-05
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-covert-racism-in-llms-b36446-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.