“The weakness in the paper is that is that is built on a bunch of benchmarks, and we all known benchmarks can be gamed.”
Our reading, the sentence that was checked:
AI benchmarks can be gamed, which weakens an AGI definition built on them.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Is AGI the right goal for AI?, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-is-agi-the-right-goal-for-ai-aaad0e-a1
We recorded 11 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-10-16. 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
- built on a bunch of benchmarks
- benchmarks can be gamed
- The weakness in the paper
Our interpretation
- The checkable core is that benchmark scores can be inflated without corresponding general capability, through contamination, overfitting or benchmark-specific optimization.
- Can expresses an established vulnerability, not an accusation that the authors actually manipulated this paper's scores.
- The period is the historical record through publication because the claim concerns a demonstrated possibility, not a periodically reported rate.
- Calling this a weakness is a methodological judgment. The assessment tests its factual premise and the resulting limitation on interpreting benchmark scores.
Counts as supported if: Primary research demonstrates benchmark-specific performance inflation that does not transfer to comparable unseen tasks.
Counts as contradicted if: Evidence establishes that the alleged inflation reflects genuine transferable capability, leaving the claimed gaming mechanism unsupported.
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
Primary research demonstrates benchmark-specific score inflation that fails to generalize, including experiments with deliberately contaminated models. This supports the factual premise that benchmarks can be gamed and limits what their scores alone establish about general competence. It does not show that the referenced AGI paper was actually gamed or that benchmark-based assessment has no value.
Limitations
- The experiments establish vulnerability, not deliberate misconduct by any particular developer.
- The degree to which this weakens the specific AGI framework remains a methodological judgment.
Evidence (1 item)
Supports the claim: arXiv; Jasper Dekoninck, Mark Niklas Müller and Martin Vechev,
data contamination can lead to inflated performance
Why it matters here: ConStat defines contamination by performance that fails to generalize to comparable tasks, tests deliberately contaminated models and identifies inflated scores in public models. This directly supports the possibility of benchmark-specific gains without matching general capability.
- Publisher
- arXiv; Jasper Dekoninck, Mark Niklas Müller and Martin Vechev
- 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, Is AGI the right goal for AI?
The weakness in the paper is that is that is built on a bunch of benchmarks, and we all known benchmarks can be gamed.
- Where
- paragraph 18
- Published
- 2025-10-16
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-is-agi-the-right-goal-for-ai-aaad0e-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-is-agi-the-right-goal-for-ai-aaad0e-a1.