“Across domains involving causality, value trade-offs, justification, and error monitoring, humans attend to reasons, stakes, and responsibility, whereas language models optimize for linguistic plausibility—structural fault lines between performance alignment and genuine evaluative competence (Quattrociocchi et al., 2025).”
Our reading, the sentence that was checked:
Across domains involving causality, value trade-offs, justification, and error monitoring, humans attend to reasons, stakes, and responsibility, whereas language models optimize for linguistic plausibility.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Rumors of AGI’s arrival have been greatly exaggerated, paragraph 28.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-rumors-of-agis-arrival-have-been-01ca07-a1
We recorded 18 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 2026-02-17. 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
- across domains involving causality, value trade-offs, justification, and error monitoring (scope, stated by the speaker)
From the speaker
- Across domains involving causality, value trade-offs, justification, and error monitoring
- humans attend to reasons, stakes, and responsibility
- language models optimize for linguistic plausibility
Our interpretation
- The ordinary reading is a broad contrast between what guides human and model judgments in the named domains.
- I distinguish a model's training objective from its observed response to evidence, consequences, and errors. One does not by itself establish the other.
- The wording does not explicitly say that linguistic plausibility is the only model objective, or that humans always reason responsibly.
- I use the cited 2025 research period through the post date for this ongoing comparison.
Counts as supported if: Controlled comparisons across the named domains show that human judgments systematically track reasons and stakes while model judgments instead track linguistic plausibility, with training and process evidence consistent with that contrast.
Counts as contradicted if: Representative comparisons show that the proposed contrast fails across the named domains, with models systematically tracking relevant reasons, consequences, and errors; training records also rule out the asserted account of their optimization.
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
The cited paper supplies a conceptual argument rather than a direct experimental demonstration of the whole contrast. Correctness-based reinforcement learning and controlled causal-reasoning results qualify a simple account in which models merely pursue plausible language. But these findings do not settle the broader comparison involving values, justification, error monitoring, and responsibility. Neither support nor contradiction of the complete claim is established.
Limitations
- A model can be trained for correctness while still making unreliable judgments.
- Success on causal tasks does not establish human-like mechanisms or responsibility.
- Treating 'optimize for linguistic plausibility' as an exclusive statement would strengthen the claim beyond its explicit wording.
Evidence (3 items)
Context only: arXiv; Quattrociocchi, Capraro, and Perc,
what fits.
Why it matters here: This Perspective develops the conceptual contrast cited in the post. Its agreement with Marcus's formulation does not independently establish that humans and models behave this way across all four named domains.
- Publisher
- arXiv; Quattrociocchi, Capraro, and Perc
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: DeepSeek-AI, hosted by arXiv,
whether the response is correct.
Why it matters here: DeepSeek documents correctness-based rewards and executable checks, so a linguistic-plausibility-only account of model training is incomplete. This does not by itself establish how models judge responsibility or value trade-offs.
- Publisher
- DeepSeek-AI, hosted by arXiv
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: arXiv; Dettki et al.,
aligned most closely with normative inferences
Why it matters here: The comparison finds substantial model sensitivity to causal structure, alongside remaining errors and differences among models. It challenges a simple human-reasoning/model-plausibility dichotomy, but covers a narrow causal task rather than the entire claim.
- Publisher
- arXiv; Dettki et al.
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Comparative studies across the named domains that separately vary relevant reasons, stakes, and surface wording, together with a clear statement of which model objectives and processes the claim concerns.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Rumors of AGI’s arrival have been greatly exaggerated
Across domains involving causality, value trade-offs, justification, and error monitoring, humans attend to reasons, stakes, and responsibility, whereas language models optimize for linguistic plausibility—structural fault lines between performance alignment and genuine evaluative competence (Quattrociocchi et al., 2025).
- Where
- paragraph 28
- Published
- 2026-02-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-rumors-of-agis-arrival-have-been-01ca07-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-rumors-of-agis-arrival-have-been-01ca07-a1.