“The paper’s core claim— “GPT-4 attains a form of general intelligence [as] demonstrated by its core mental capabilities (such as reasoning, creativity, and deduction)”—literally cannot be tested with serious scrutiny, because the scientific community has no access to the training data.”
Our reading, the sentence that was checked:
The Sparks of AGI paper's claim that GPT-4 attains a form of general intelligence cannot be seriously tested because scientists have no access to its training data.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The Sparks of AGI? Or the End of Science?, paragraph 21.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-sparks-of-agi-or-the-end-of-science-3a62a4-a1
We recorded 7 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 2023-03-24. 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)
Conditions
- because the training data is not accessible (qualifier, stated by the speaker)
From the speaker
- The paper’s core claim
- a form of general intelligence
- literally cannot be tested with serious scrutiny
- because the scientific community has no access to the training data
Our interpretation
- Read as a necessity claim: training-data secrecy prevents serious empirical testing of the claimed general capabilities, rather than merely making contamination audits harder.
- The target remains the early GPT-4 model and intelligence claim in Sparks. Later research can inform the methodological question, but tests of another checkpoint are context.
- Serious scrutiny means controlled, reproducible tests capable of challenging the claim; it does not require conclusively proving AGI.
Counts as supported if: Methodological evidence establishes that access to the training corpus is indispensable for meaningful tests of the paper's claimed general capabilities.
Counts as contradicted if: Rigorous tests of the relevant model's claimed general capabilities demonstrably work without access to its training corpus.
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
Independent counterfactual tests show that researchers could meaningfully investigate GPT-4's reasoning without pretraining details. That weighs against a broad claim that secrecy makes scrutiny impossible. However, the study concerns the released checkpoint and selected capabilities, not a direct test of the exact early model and the paper's overall intelligence claim. The record does not settle the causal assertion as scoped.
Limitations
- This is not support for the claim that all serious testing requires training-data access.
- A broader reading about the possibility of behavioral scrutiny faces substantial counterevidence.
- The distinction between the exact early model and the release checkpoint materially affects the result.
Evidence (1 item)
Context only: MIT and Boston University researchers via arXiv,
underperforms on counterfactual variants compared to default task instantiations.
Why it matters here: Researchers tested transfer across altered task rules using gpt-4-0314 without pretraining details, demonstrating meaningful scrutiny of reasoning. This is material counterevidence to a broad impossibility claim, but it tests the released checkpoint rather than the paper's exact early model.
- Publisher
- MIT and Boston University researchers via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
Independent, controlled tests of the relevant early model that address the paper's stated general capabilities without corpus access, or a defensible methodological demonstration that such tests require that access.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, The Sparks of AGI? Or the End of Science?
The paper’s core claim— “GPT-4 attains a form of general intelligence [as] demonstrated by its core mental capabilities (such as reasoning, creativity, and deduction)”—literally cannot be tested with serious scrutiny, because the scientific community has no access to the training data.
- Where
- paragraph 21
- Published
- 2023-03-24
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-sparks-of-agi-or-the-end-of-science-3a62a4-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-the-sparks-of-agi-or-the-end-of-science-3a62a4-a1.