“LLMs are already dangerous, despite their limits.”
Our reading, the sentence that was checked:
LLMs are already dangerous despite their limitations.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Hard-forked! Casey Newton’s distorted portrait of Gary Marcus and AI skepticism, paragraph 26.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-hard-forked-casey-newtons-distorted-808ee9-a3
We recorded 14 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 2024-12-06. 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
- despite their limits (qualifier, stated by the speaker)
From the speaker
- LLMs are already dangerous
- despite their limits
- Covert racism?
- Propaganda?
Our interpretation
- The plain reading is that existing LLMs pose concrete risks of material harm, including discriminatory behavior and assistance to deceptive influence operations.
- Dangerous does not require proof that every use is harmful or that every example in the surrounding list has already caused documented injury.
- Despite their limits acknowledges that limited capability and danger can coexist; it is not a conditional forecast.
- Because already refers to dangers present by publication and the claim has no reporting cycle, use calendar year 2024 through December 6 as the current observation window.
- The separate suggestion of imminent military misuse is outside the proposition being graded.
Counts as supported if: Evidence available by publication demonstrates materially discriminatory LLM behavior in relevant evaluations or actual use of LLMs to assist harmful deception.
Counts as contradicted if: Evidence establishes that the alleged risk mechanisms are absent or effectively prevented across the relevant systems and uses. Low impact in particular incidents alone does not disprove the existence of danger.
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
A March 2024 preprint found that tested language models assigned less prestigious jobs to speakers of African American English and selected the death penalty more often in hypothetical judgments involving that dialect [e1]. OpenAI's May 30, 2024 report documented model use in covert influence operations, while finding no meaningful increase in those campaigns' reach or engagement attributable to its services as of May [e2]. These findings support the existence of discriminatory risks and assistance to deceptive activity before the December 6 claim, without establishing downstream injury [e1] [e2].
Limitations
- The experiments do not establish actual hiring decisions or criminal sentences [e1].
- The influence-operation investigation does not establish changes in public opinion or political outcomes [e2].
- These sources do not establish how frequent or severe harms were across deployments, or substantiate the surrounding examples involving insurance, housing and military misuse [e1] [e2].
Evidence (2 items)
Supports the claim: arXiv,
dialect prejudice
Why it matters here: The researchers demonstrated unfavorable job assignments and harsher hypothetical criminal judgments associated with African American English. This establishes a concrete discriminatory risk, while the experiment does not establish that employers or courts actually imposed these outcomes.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: OpenAI,
covert influence operations
Why it matters here: OpenAI documented operations using its models to generate political comments and articles that were subsequently posted online. This establishes actual assistance to deceptive activity, but its investigation found no meaningful increase in audience reach or engagement attributable to its services.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Hard-forked! Casey Newton’s distorted portrait of Gary Marcus and AI skepticism
LLMs are already dangerous, despite their limits.
- Where
- paragraph 26
- Published
- 2024-12-06
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-hard-forked-casey-newtons-distorted-808ee9-a3: 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-hard-forked-casey-newtons-distorted-808ee9-a3.
HeldBy this site
The final-text check failed: citations_correct.
About assessment marcus-hard-forked-casey-newtons-distorted-808ee9-a2, which was not published.