“All that really happened is that GPT-5 (oversimplifying slightly) crawled the web for solutions to already-solved problems.”
Our reading, the sentence that was checked:
GPT-5 did not solve open Erdős problems; it essentially located existing solutions to already-solved problems on the web.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Erdosgate, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-erdosgate-a597e8-a1
We recorded 5 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 2025-10-21. 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
- Erdős problems episode (qualifier, stated by the speaker)
From the speaker
- All that really happened
- GPT-5 (oversimplifying slightly)
- crawled the web for solutions to already-solved problems
Our interpretation
- The ordinary reading concerns the essential result of this episode: locating existing mathematical solutions rather than resolving genuinely open conjectures.
- The express qualification allows assistance with reading, translation, explanation, and human verification; it does not assert a literal crawler-only implementation.
- The period covers the October 11 antecedent post identified in the supplied update through the October 21 article.
Counts as supported if: First-hand records establish that the announced solutions came from existing literature rather than new resolutions of previously unsolved conjectures.
Counts as contradicted if: At least one result in the announced episode is established as a genuinely new resolution of a previously unsolved conjecture.
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 first-hand explanation and dated mathematical records support the central claim: this episode located existing solutions rather than resolving previously unsolved conjectures. GPT-5 also translated and explained material. Marcus's explicit allowance for oversimplification makes the central reading supportable, although a literal claim that nothing beyond retrieval occurred would be too strong.
Limitations
- The result depends on treating the crawler wording as the acknowledged simplification.
- This assessment does not address later AI mathematical discoveries.
Evidence (3 items)
Supports the claim: Sébastien Bubeck, X,
it’s not about AIs discovering new results on their own
Why it matters here: Bubeck directly characterizes the episode as work with existing knowledge. His detailed example identifies an earlier published solution rather than a newly proved result.
- Publisher
- Sébastien Bubeck, X
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: Mehtaab Sawhney, Erdős Problems forum,
was located by GPTPro
Why it matters here: The comment identifies a 2003 paper and its Theorem 4 as solving the problem, then credits the model with locating the reference. The event date records that discovery, not the original mathematical result.
- Publisher
- Mehtaab Sawhney, Erdős Problems forum
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Mark Sellke, Erdős Problems forum,
translated
Why it matters here: Sellke identifies earlier papers as the solution and describes translation and explanation by GPT-5. This qualifies a literal reading that the model merely returned links, without establishing a new mathematical result.
- Publisher
- Mark Sellke, Erdős Problems forum
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Erdosgate
All that really happened is that GPT-5 (oversimplifying slightly) crawled the web for solutions to already-solved problems.
- Where
- paragraph 4
- Published
- 2025-10-21
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-erdosgate-a597e8-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.