“Because both countries are both building essentially the same kind of models with the same kinds of techniques using the same kinds of data, the results from the two nations are essentially converging on the same outcomes.”
Our reading, the sentence that was checked:
US and Chinese AI results are converging because both use the same kinds of models, techniques, and data.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, When it Comes to AI Policy, Congress Shouldn’t Cut States off at the Knees, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-when-it-comes-to-ai-policy-congress-013148-a1
We recorded 15 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 2025-05-14. 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)
From the speaker
- Because
- both countries
- essentially the same kind of models
- the same kinds of techniques
- the same kinds of data
- essentially converging on the same outcomes
Our interpretation
- Read the leaderboard context as referring to leading generative language and reasoning models, not every AI application.
- 'Same kinds' means broad similarities, not identical architectures, training recipes, or datasets.
- Assess convergence during 2024 through the post date, using the latest completed annual period and the subsequent prepublication model releases. The speaker does not specify the starting date.
- Preserve the causal commitment: narrowing benchmark gaps and shared methods alone do not establish that those similarities explain the narrowing.
Counts as supported if: Comparable evaluations show narrowing US–China capability gaps, and credible causal or comparative evidence connects that narrowing to shared model types, training methods, and data types.
Counts as contradicted if: Comparable evidence shows nonconvergence over the described period, or evidence directly undermines the proposed causal explanation.
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
The convergence component has support: Stanford reports a narrowing US–China performance gap, and the original model papers document reinforcement learning on both sides. The stronger 'because' claim remains unsettled. These sources do not show that similarities in models, methods, and data caused the narrowing. Research on correlated errors also cautions that similar behavior can occur with different architectures. That caution is not a direct refutation of Marcus's explanation.
Limitations
- A supported assessment of benchmark convergence alone would drop the claim's causal commitment.
- Two model families cannot establish uniformity across either country's AI development.
- The contextual error-correlation study measures a different outcome and is not decisive counterevidence.
Evidence (4 items)
Supports the claim: Stanford Institute for Human-Centered Artificial Intelligence,
The U.S. still leads in producing top AI models—but China is closing the performance gap.
Why it matters here: The accompanying paragraph reports that MMLU and HumanEval gaps narrowed from double digits in 2023 to near parity in 2024. This supports the convergence component, but does not identify its cause.
- Publisher
- Stanford Institute for Human-Centered Artificial Intelligence
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: DeepSeek-AI, arXiv,
Through RL, DeepSeek-R1-Zero naturally emerges with numerous powerful and intriguing reasoning behaviors.
Why it matters here: Documents reinforcement learning as a mechanism used by a leading Chinese developer. Together with the o1 system card it establishes one broad methodological similarity, not a country-level causal explanation.
- Publisher
- DeepSeek-AI, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI, arXiv,
The o1 model series is trained with large-scale reinforcement learning to reason using chain of thought.
Why it matters here: Establishes that a leading US model family also used reinforcement learning for reasoning. It does not establish equivalent data, identical training procedures, or why the national performance gap narrowed.
- Publisher
- OpenAI, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Proceedings of Machine Learning Research,
Crucially, however, larger and more accurate models have highly correlated errors, even with distinct architectures and providers.
Why it matters here: Shows why similar model behavior cannot by itself establish a shared-architecture explanation. This is a caution about inference, not direct contradiction: correlated errors differ from national capability convergence, and the study is not matched to the claim's exact period.
- Publisher
- Proceedings of Machine Learning Research
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
What would settle it
A comparative study that measures the narrowing gap and tests how much shared architectures, methods, and data explain it, while accounting for alternative explanations and evaluation differences.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, When it Comes to AI Policy, Congress Shouldn’t Cut States off at the Knees
Because both countries are both building essentially the same kind of models with the same kinds of techniques using the same kinds of data, the results from the two nations are essentially converging on the same outcomes.
- Where
- paragraph 9
- Published
- 2025-05-14
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-when-it-comes-to-ai-policy-congress-013148-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-when-it-comes-to-ai-policy-congress-013148-a1.