“This is even more true four years later, now that we have systems that ingest massive amounts of data from both public and commercial sources, quite possibly including eg sample law school and medical exams etc.”
Our reading, the sentence that was checked:
Because current systems ingest massive data possibly including sample exams, benchmark success is even less indicative of intelligence than in 2019.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, No, Virginia, AGI is not imminent, paragraph 11.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-no-virginia-agi-is-not-imminent-27976b-a1
We recorded 17 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 2023-08-20. 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
- even more true four years later
- ingest massive amounts of data
- both public and commercial sources
- quite possibly including eg sample law school and medical exams
Our interpretation
- The claim concerns a deterioration between 2019 and August 20, 2023, caused by expanded training-data exposure.
- I interpret benchmark informativeness as how well scores predict performance on genuinely unseen tasks. This is an operational measure, not a definition of intelligence.
- Possible exam exposure is a proposed mechanism, not a firm assertion that particular exams were included.
Counts as supported if: Comparable evaluations of 2019 and pre-post 2023 systems show that expanded training exposure caused benchmark scores to become less predictive of unseen-task performance.
Counts as contradicted if: Comparable evaluations show that expanded exposure did not reduce benchmark predictive validity, or that validity improved after controlling for other changes.
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
Contamination can inflate scores, but that does not establish that benchmarks became less informative between 2019 and August 2023 because training corpora expanded. The controlled study distinguishes memorization from exploitation, and OpenAI reported little contamination effect on its exam results. Neither supplies the comparison needed to settle Marcus's causal claim.
Limitations
- OpenAI's audit was a developer assessment with acknowledged detection limits.
- Unseen-task generalization captures only part of what intelligence might mean.
Evidence (2 items)
Context only: arXiv; Inbal Magar and Roy Schwartz,
exploit the contaminated data
Why it matters here: Controlled experiments found that contamination could improve downstream results, but memorization did not always produce that improvement. This establishes a possible mechanism without measuring a deterioration from 2019 to 2023.
- Publisher
- arXiv; Inbal Magar and Roy Schwartz
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
very little effect
Why it matters here: OpenAI reported little change after removing detected contaminated exam questions. Its audit acknowledged detection limitations and did not provide the longitudinal comparison needed to contradict the full claim.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
A controlled comparison of 2019 and pre-August 20, 2023 systems measuring benchmark predictive validity with and without training exposure to benchmark material.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, No, Virginia, AGI is not imminent
This is even more true four years later, now that we have systems that ingest massive amounts of data from both public and commercial sources, quite possibly including eg sample law school and medical exams etc.
- Where
- paragraph 11
- Published
- 2023-08-20
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-no-virginia-agi-is-not-imminent-27976b-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-no-virginia-agi-is-not-imminent-27976b-a1.