“In our view, inadequate attention to these three factors has serious consequences, including:”
Our reading, the sentence that was checked:
Neglecting reference, cognitive models and compositionality causes serious problems in large language models.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Three ideas from linguistics that everyone in AI should know, paragraph 8.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-three-ideas-from-linguistics-that-0eeecb-a1
We recorded 26 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 2022-06-22. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- inadequate attention to these three factors
- has serious consequences
- Reference
- Cognitive models
- Compositionality
- lose coherence over time
- distinguishing truth from falsehoods
- bias and toxic speech
Our interpretation
- Read in context, this attributes a substantial causal contribution to inadequate reference, cognitive models and compositionality across the three listed problems.
- It does not say that each factor independently causes every problem, or that these are the only causes.
- Operational interventions may approximate the linguistic concepts, but that correspondence must be justified rather than assumed.
- The claim concerns models available at the post, not a forecast about subsequent models.
Counts as supported if: Controlled interventions or comparably strong causal evidence connect the specified deficits to substantial coherence, truthfulness and harmful-speech problems, with suitable controls for other changes.
Counts as contradicted if: Comparable causal tests establish that the specified deficits do not materially contribute to the listed problems, with enough coverage to challenge the general attribution.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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 research makes parts of the explanation plausible, but does not settle the broad causal attribution. LaMDA improves with external knowledge and targeted training, while other work finds useful implicit state representations in text-trained models. Neither result establishes how inadequate reference, cognitive models and compositionality contribute across all three consequences. Evidence that the problems exist cannot by itself establish this explanation.
Limitations
- The search is not an exhaustive systematic review.
- Retrieval-based groundedness is an imperfect operational substitute for linguistic reference.
- Evidence from BART and T5 does not automatically establish mechanisms in GPT-3 or LaMDA.
Evidence (2 items)
Context only: Google; arXiv,
access external knowledge sources.
Why it matters here: LaMDA comparisons connect external knowledge access with better groundedness and safety-specific training with better safety. They provide relevant intervention evidence, but neither isolate Marcus’s three concepts nor test all three claimed consequences.
- Publisher
- Google; arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Belinda Z. Li, Maxwell Nye and Jacob Andreas,
implicit models are approximate and error-prone.
Why it matters here: The researchers recover and manipulate entity-state representations in BART and T5, while acknowledging their errors. This complicates an absence-based explanation but does not determine whether inadequate versions of those capabilities cause the three listed problems.
- Publisher
- arXiv; Belinda Z. Li, Maxwell Nye and Jacob Andreas
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Controlled studies on the historical model class that define and vary the proposed capabilities, measure all three consequences, and separate their effects from supervision, scale and decoding changes.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Three ideas from linguistics that everyone in AI should know
In our view, inadequate attention to these three factors has serious consequences, including:
- Where
- paragraph 8
- Published
- 2022-06-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-three-ideas-from-linguistics-that-0eeecb-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-three-ideas-from-linguistics-that-0eeecb-a1.