“Bing relies on GPT-4, and we (eg, in the scientific community) don’t have access to how GPT-4 works, and we don’t have access to what data its trained on (vital, since we know that systems can bias eg political thought and hiring decisions based on those undisclosed data) — that’s about as far away from transparency as we could be.”
Our reading, the sentence that was checked:
Scientists lack access to how GPT-4 works and its training data, making it about as far from transparent as possible.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Two models of AI oversight - and how things could go deeply wrong, paragraph 18.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-two-models-of-ai-oversight-and-how-7164b6-a2
We recorded 20 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 2023-06-08. 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
- Bing relies on GPT-4
- we (eg, in the scientific community) don’t have access to how GPT-4 works
- we don’t have access to what data its trained on
- about as far away from transparency as we could be
Our interpretation
- The checkable core concerns public scientific access to GPT-4's detailed architecture, training process, and training dataset.
- Knowing the general model family or being able to submit prompts does not constitute access to those details.
- The transparency superlative is read as rhetorical emphasis on substantial opacity, not a measurable ranking of every possible disclosure policy.
- The parenthetical discussion of political and hiring bias explains why transparency matters; it is not treated as a separate claim that GPT-4 caused a particular discriminatory decision.
- There is no regular disclosure reporting period. I use the current GPT-4 release and documentation cycle, beginning in March 2023 and ending at the post. March 1 denotes the beginning of that month, not an asserted release day.
Counts as supported if: The contemporaneous disclosure record withholds detailed architecture and training-dataset information from the general scientific community, despite allowing use or high-level study of the model.
Counts as contradicted if: Contemporaneous records establish that the scientific community could inspect detailed architecture and the training dataset, rather than only model outputs or broad descriptions.
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 March 27, 2023 technical report discloses broad principles but withholds detailed architecture and dataset-construction information [e2]. The March 30 CAIDP filing also reports these disclosure gaps [e1]. This supports the claim of substantial public opacity during the March–June 8, 2023 observation window [e1, e2]. It does not imply that researchers knew nothing: the report identifies GPT-4 as a Transformer trained to predict the next token [e2].
Limitations
- The evidence does not establish every confidential research-access arrangement.
- No defined transparency scale establishes the literal superlative.
- The CAIDP filing is an advocacy document, not a regulatory finding [e1].
Evidence (2 items)
Supports the claim: Center for AI and Digital Policy,
OpenAI
Why it matters here: The contemporaneous complaint identifies the specific missing disclosures. Its statement is corroborated by OpenAI's own report; the complaint itself is not an adjudicated regulatory finding.
- Where
- Paragraph 108
- Publisher
- Center for AI and Digital Policy
- Kind of source
- Regulatory filing
- Describes events of
- Public from
Supports the claim: OpenAI / arXiv,
Transformer-based model
Why it matters here: Section 2 explicitly withholds detailed architecture, model size, compute, dataset construction, and training-method information. The quoted abstract and Section 2 also disclose broad principles, data-source categories, and reinforcement learning from human feedback, qualifying any literal suggestion that nothing was known.
- Publisher
- OpenAI / arXiv
- 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, Two models of AI oversight - and how things could go deeply wrong
Bing relies on GPT-4, and we (eg, in the scientific community) don’t have access to how GPT-4 works, and we don’t have access to what data its trained on (vital, since we know that systems can bias eg political thought and hiring decisions based on those undisclosed data) — that’s about as far away from transparency as we could be.
- Where
- paragraph 18
- Published
- 2023-06-08
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-two-models-of-ai-oversight-and-how-7164b6-a2: 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-two-models-of-ai-oversight-and-how-7164b6-a2.