With medium confidence, technical moat will continue to be elusive in 2025, with convergence on broadly similar models across the US and China and some European systems catching up to roughly the same place.
Unresolvable The public record could not settle it.
Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?
The second judge would not resolve this claim; the verdict shown is the first judge's.
What would count as true or false
Judged at the end of 2025. The claim, as glossed in the proposition, has three parts: (a) no AI developer held a lasting technical lead (moat) during 2025; (b) US and Chinese models converged on similar capability; (c) some European systems caught up. True: 2025 leaderboards and reviews show top models from several US labs and Chinese labs (e.g. DeepSeek, Qwen) within a small gap of each other, leads that changed hands or were quickly matched, and at least one European system (e.g. Mistral) near the frontier on standard benchmarks. Mostly true: (a) and (b) hold but the European catch-up was only partial. Mixed: one of (a) or (b) holds and the other does not, e.g. convergence but one lab clearly and persistently ahead. Mostly false: a single developer kept a clear, widely recognised lead through 2025, with only partial convergence elsewhere. False: one company or country built a durable, large capability lead that others did not close during 2025.
Set before the research began.
Evidence (4)
-
E1 Technical Performance Stanford Institute for Human-Centered AI (AI Index Report 2026, Chapter 2) · 16 August 2026
“In February 2025, DeepSeek-R1 (1,400) trailed the leading U.S. model (o1-2024-12-17, 1,405) by just 5 Arena points (0.4%).”
-
E2 ai-model-arena-analysis/analysis/FINDINGS.md at main · hashtro9-rgb/ai-model-arena-analysis GitHub (anonymous user repository hashtro9-rgb) · 14 July 2026
“Google's breakout: a single unbroken 47-week reign (Jun 2025 -> Jan 2026) — every one of Google's #1 finishes came in that one streak.”
-
E3 ai-model-arena-analysis/analysis/FINDINGS.md at main · hashtro9-rgb/ai-model-arena-analysis GitHub (anonymous user repository hashtro9-rgb) · 14 July 2026
“no single lab has a durable moat; a competitor's next release routinely erases a leader's edge within weeks.”
-
E4 US AI Models Lead China by 7 Months: No Chinese AI Has Yet Matched OpenAI's o3 Trending Topics · 9 January 2026
“Chinese models reach the capabilities of their US counterparts with an average delay of seven months, as an analysis of the Epoch Capabilities Index (ECI) shows.”
Reasoning
On part (a), the evidence conflicts. E3 says no lab has a durable moat, but E2, from the same anonymous analysis, reports a single unbroken 47-week reign by Google from June 2025 to January 2026. On part (b), E1 shows a 5-point US-China gap in February 2025, while E4 reports an average seven-month lag, and no evidence covers European systems, so the evidence does not settle which verdict band under the criteria applies.
Cite this claim
Gary Marcus, “With medium confidence, technical moat will continue to be elusive in 2025, with convergence on broadly similar models across the US and China and some European systems catching up to roughly the same place.” (said December 2024). Verdict: unresolvable. Reliability API, claim c_134dad1ed79a, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_134dad1ed79a
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More prediction claims by Gary Marcus
-
Resolved true
“Profits from AI models will continue to be modest or nonexistent”
Profits from AI models will remain modest or nonexistent in 2025.
-
Mostly true
“GPT-5 may be more satisfying but is likely even more expensive to train and operate.”
GPT-5 is likely to be even more expensive to train and operate than GPT-4.
-
Mostly true
“Some of its advice will be head-scratchingly bad.”
-
Mostly true
“GPT-5 will surely be better, a lot better than GPT-4.”
-
Resolved true
“GPT-5 really is about to drop, no foolin’.”
Verdicts are made automatically; how verdicts are made.