“We know nothing about the architecture.”
Our reading:
Nothing is publicly known about the architecture of the OpenAI system that produced the math results.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
OpenAI’s October 6, 2026 documentation identifies Lean formalizations of some released results [e3]. Those artifacts alone do not establish how the generating system connected proof generation, iteration and verification [e3]. Resolving the claim requires evidence about the system’s architecture and what was public by Marcus’s October 7 publication.
Caveats
- Published formalizations do not establish whether generation used Lean feedback or formalization occurred afterward.
- The supplied evidence does not establish that no substantive architectural information was public during the target period.
Evidence (1 item)
Context only: OpenAI,
Lean formalizations
Why it matters: Lean artifacts were public in the initial release. Their existence does not establish whether generation used Lean feedback, whether formalization happened afterward, or whether the process was iterative.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- Marcus's examples make architecture include the arrangement of generation, iteration and verification components, not just the neural network's internal design.
- Nothing means no substantive architectural information about the relevant system. It does not require disclosure of weights or every implementation detail.
- Assess the public disclosure state from the October 6 release through publication on October 7.
Right if: The contemporaneous record discloses no substantive information about the relevant model architecture or generation-and-verification arrangement.
Wrong if: A contemporaneous source identifies at least one substantive architectural feature and establishes that it belongs to the system used for these results.
We wrote this test after the outcome was known.
What would settle it
A disclosure available by Marcus’s October 7 publication that identifies a substantive architectural feature of the system used for these results, or a sufficiently complete contemporaneous record establishing that no such information was public.
We found 7 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Brief remark on OpenAI’s giant math drop, paragraph 14.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2026-10-07.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Brief remark on OpenAI’s giant math drop
We know nothing about the architecture.
- Where
- paragraph 14
- Published
- 2026-10-07
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
The final check failed: a cited source didn’t say what the write-up said.