“Since then, the results on benchmarks have gotten better, but there’s still something missing.”
Our reading, the sentence that was checked:
Since 2019 large language models have done better on benchmarks, but something is still missing from them.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Three ideas from linguistics that everyone in AI should know, paragraph 3.
- 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-5c8654-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
- 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)
From the speaker
- Since then
- the results on benchmarks have gotten better
- there’s still something missing
Our interpretation
- The comparison starts in 2019, the year identified immediately before this sentence, and ends at publication.
- The plain reading is that benchmark performance improved while meaningful limitations remained. It does not require every benchmark to improve.
- The surrounding discussion makes factual reliability, grounding and compositional understanding relevant limitations. Supporting this broad sentence does not establish every stronger claim that follows.
Counts as supported if: Published benchmark comparisons show improvement after 2019, alongside evaluations of contemporary models showing a substantial remaining limitation identified in the surrounding discussion.
Counts as contradicted if: Comparable results show no benchmark improvement, or evaluations establish that the relevant remaining limitations had been resolved by publication.
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
The broad observation holds. PaLM documented continuing benchmark gains, while TruthfulQA demonstrated substantial factual unreliability in contemporary language models, including GPT-3. This supports improvement alongside remaining limitations. It does not validate every stronger explanation of those limitations in the surrounding text.
Limitations
- TruthfulQA deliberately tests susceptibility to misconceptions; its percentages are not general error rates for all language use.
- The evidence covers particular models and evaluations, not every model or benchmark.
- The result depends on reading this sentence as a broad observation rather than a conjunction of the three stronger claims that follow.
Evidence (2 items)
Supports the claim: arXiv; Chowdhery et al.,
state-of-the-art few-shot learning results
Why it matters here: PaLM's contemporary report documents continued benchmark improvements. This supports the improvement component, without establishing that benchmark success implies general understanding.
- Where
- Abstract
- Publisher
- arXiv; Chowdhery et al.
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: arXiv; Lin, Hilton and Evans,
truthful on 58% of questions
Why it matters here: The evaluation included GPT-3 and found substantial factual unreliability on questions designed to elicit common misconceptions. This supports a remaining limitation specifically discussed in the post.
- Where
- Abstract
- Publisher
- arXiv; Lin, Hilton and Evans
- 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, Three ideas from linguistics that everyone in AI should know
Since then, the results on benchmarks have gotten better, but there’s still something missing.
- Where
- paragraph 3
- Published
- 2022-06-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-three-ideas-from-linguistics-that-5c8654-a1: 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-three-ideas-from-linguistics-that-5c8654-a1.