“And I still attribute those problems to a lack of world models that represent the ways in which things unfold over time. I still see this as the central flaw in large language models.”
Our reading, the sentence that was checked:
The dopey outputs of GPT-style LLMs are caused by a lack of world models of how things unfold over time, which is the central flaw of LLMs.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Gaslighting and reality in AI, paragraph 28.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gaslighting-and-reality-in-ai-77c9cb-a1
We recorded 4 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 2023-02-02. 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
- attribute those problems to a lack of world models
- represent the ways in which things unfold over time
- the central flaw in large language models
Our interpretation
- Those problems refers to the nonsensical outputs and elementary reasoning failures discussed in the surrounding passage.
- The plain reading of central flaw is a principal explanation of these failures, not merely one contributing factor.
- The claim concerns the GPT-style models discussed from 2019 through the post date, rather than models released afterward.
Counts as supported if: Controlled interventions show that missing or defective temporal state representations cause the relevant failures, and that this mechanism explains a principal share of them compared with plausible alternatives.
Counts as contradicted if: Controlled studies show that the relevant models already possess adequate temporal representations and that other mechanisms principally cause the cited failures, or that repairing temporal representations does not materially address them.
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
Controlled research supports state reasoning as a contributing factor: supplying entity-state information improved coherence on two tasks. That is weaker than showing that missing world models caused the cited mistakes or constituted the central flaw of LLMs. The retrieved studies do not compare this explanation with other causes across a representative set of failures.
Limitations
- Task-specific improvements do not establish a general causal ranking.
- The intervention does not cleanly distinguish absent representations from inaccurate or poorly used representations.
Evidence (1 item)
Supports the claim: arXiv; Belinda Z. Li, Maxwell Nye, and Jacob Andreas,
we use entity states as auxiliary supervision to improve LMs ability to perform this reasoning under both fine-tuning and prompting.
Why it matters here: Providing state supervision improves performance on TextWorld and TRIP, supporting a causal contribution from state reasoning. The experiments neither establish that representations were absent nor rank this mechanism as the principal cause of general GPT failures.
- 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 interventions on the relevant model versions and errors, comparing temporal-state repairs with other repairs and measuring which mechanisms explain most failures.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Gaslighting and reality in AI
And I still attribute those problems to a lack of world models that represent the ways in which things unfold over time. I still see this as the central flaw in large language models.
- Where
- paragraph 28
- Published
- 2023-02-02
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gaslighting-and-reality-in-ai-77c9cb-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-gaslighting-and-reality-in-ai-77c9cb-a1.