“Same story; only the architectures and training regimes have changed. The troubles with abstraction persist.”
Our reading, the sentence that was checked:
LLMs doing arithmetic still have trouble with abstraction, even with new architectures and training regimes.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Happy Groundhog Day, The AI Edition, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-happy-groundhog-day-the-ai-edition-f7b173-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 2023-02-03. 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
- in arithmetic (qualifier, stated by the speaker)
From the speaker
- Yasaman Razeghi’s analysis of LLMs doing arithmetic
- from last year
- The troubles with abstraction persist
Our interpretation
- Last year refers to 2022.
- The ordinary reading is that the arithmetic evaluations still reveal unreliable generalization across number cases despite newer architectures and training.
- Only the architectures and training regimes have changed is read as rhetorical emphasis on persistence, not a literal denial of every other scientific change.
- Training-frequency dependence is evidence of limited robustness; it is not proof that a model possesses no abstraction ability.
Counts as supported if: The cited 2022 study finds material arithmetic-performance differences associated with the familiarity of numbers or their combinations in pretraining.
Counts as contradicted if: The cited study instead establishes robust arithmetic generalization across those cases without the alleged familiarity dependence.
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
Razeghi and colleagues' study, in the version available May 24, 2022, found substantial addition and multiplication performance gaps associated with pretraining frequency in the tested GPT-based models [e1]. This supports continuing difficulty with reliable arithmetic generalization in those models [e1]. The study also reports a simple task with a very small performance gap and discusses possible generalization, limiting any broader conclusion that the models cannot generalize at all [e1].
Limitations
- Frequency-associated performance gaps do not prove that models lack all abstraction ability or rely entirely on memorization [e1].
- The findings concern the selected models and prompting conditions studied in 2022; they do not establish performance across all LLMs or in 2026 [e1].
Evidence (1 item)
Supports the claim: arXiv,
strong effect of frequency
Why it matters here: The study finds substantial frequency-associated gaps on addition and multiplication in the tested GPT-based models. This supports continuing difficulty with reliable arithmetic generalization, without establishing that all performance is memorization.
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Happy Groundhog Day, The AI Edition
Same story; only the architectures and training regimes have changed. The troubles with abstraction persist.
- Where
- paragraph 9
- Published
- 2023-02-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-happy-groundhog-day-the-ai-edition-f7b173-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.
About assessment marcus-happy-groundhog-day-the-ai-edition-f7b173-a2.