“Almost nothing is known about the precise internal mechanisms of GPT-4 or what it has trained on.”
Our reading, the sentence that was checked:
Almost nothing is known about the precise internal mechanisms of GPT-4 or what it has been trained on.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, OpenAI’s “Own Goal”, paragraph 4.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-openais-own-goal-3c5406-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 2024-03-06. 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
- almost nothing (scope, stated by the speaker)
- precise internal mechanisms (scope, stated by the speaker)
From the speaker
- Almost nothing is known
- precise internal mechanisms of GPT-4
- what it has trained on
Our interpretation
- The plain reading concerns reliable public knowledge of implementation details and training-data composition, rather than private knowledge inside OpenAI.
- Almost nothing allows broad descriptions of the model family and training process; it does not mean literally no information.
- Use the latest technical-report snapshot public before the post, revised March 4, 2024, through March 6, 2024.
- Consider independent reporting as well as official disclosures, but distinguish reported specifications from publicly verifiable specifications.
Counts as supported if: Public records leave the detailed architecture, model size, training setup, and training-corpus composition substantially undisclosed, despite broad descriptions.
Counts as contradicted if: Reliable public documentation or independently substantiated research provides substantial precise knowledge of GPT-4's implementation and training corpus.
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
Supported on the reading of reliable, publicly verifiable technical detail. The latest report expressly withheld the key implementation and dataset specifications while supplying broad descriptions. However, architecture reporting already existed, including Patel's mixture-of-experts account. The claim would be too sweeping if read as saying that no meaningful technical information had been reported.
Limitations
- Official nondisclosure does not prove that outsiders knew nothing.
- The result depends on distinguishing substantiated public knowledge from reported but unverified specifications.
- The quoted sentence identifies the report's scope; the detailed exclusions are in the remainder of the cited paragraph.
Evidence (2 items)
Supports the claim: OpenAI,
capabilities, limitations, and safety properties
Why it matters here: The same paragraph expressly withholds detailed architecture, size, hardware, compute, dataset construction, and training methods. It nevertheless identifies a Transformer-style model, next-token pretraining, public and licensed data, and reinforcement learning from human feedback.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Semafor,
GPT-4 has 16 experts.
Why it matters here: Patel publicly reported a specific architecture and discussed its operation. This qualifies the claim about public knowledge, but the interview does not expose evidence independently establishing the specification; it also predates the target snapshot.
- Publisher
- Semafor
- Kind of source
- News report
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, OpenAI’s “Own Goal”
Almost nothing is known about the precise internal mechanisms of GPT-4 or what it has trained on.
- Where
- paragraph 4
- Published
- 2024-03-06
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-openais-own-goal-3c5406-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.