By Gary Marcus's own count, sixteen of his seventeen high-confidence predictions about 2025 proved correct.

Judge as of January 2026. Identify the 17 'high confidence' predictions about 2025 that Marcus published (likely in his Substack around the end of 2024), and check each against what actually happened in 2025, without relying on his own scoring alone. True: an independent assessment finds 16 of 17 (or 17) correct. Mostly true: 14–15 are correct, or 16 are only with lenient readings of some vague predictions. Mixed: about 9–13 are correct. Mostly false: about 5–8 are correct. False: fewer than 5 are correct, or there was no such list of 17 high-confidence predictions.

Set before the research began.

1. E1 2025 in AI predictions — LessWrong LessWrong (post by jessicata) · 2 January 2026

   “Gary Marcus*: “No single system will solve more than 4 of the AI 2027 Marcus-Brundage tasks by the end of 2025. I wouldn’t be shocked if none were reliably solved by the end of the year.” Evaluation: Correct. AI can perhaps pass the reading comprehension task. But not any 4 of the tasks.”

2. E2 AI Policy News aipolicy.onair.cc (AI Policy News aggregator) · 20 December 2025 Own record

   “AGI didn’t materialize (contra predictions from Elon Musk and others); GPT-5 was underwhelming, and didn’t solve hallucinations.”

3. E3 25 AI Predictions for 2025, from Marcus on AI Marcus on AI (Gary Marcus Substack) · 1 January 2025 Own record

   “here my own predictions for where we will be at the end of this year:”

4. E4 GPT-5 will blow minds — Gary Marcus Prediction Scorecard · 27 July 2025

   “GPT-5 (7 August 2025) was much better than the original GPT-4 on benchmarks such as SWE-bench (74.9% vs GPT-4o's 30.8%), but its launch was widely received as underwhelming, with a user backlash that made OpenAI restore GPT-4o; Marcus himself called it "underwhelming" in December 2025.”

5. E5 2025 in AI predictions — LessWrong LessWrong · 2 January 2026

   “Evaluation: Correct. AI can perhaps pass the reading comprehension task. But not any 4 of the tasks.”

6. E6 AI Policy News aipolicy.onair.cc (aggregator page reproducing a Marcus on AI post dated 2025-12-20) · 20 December 2025 Own record

   “No system solved more than 4 (or maybe any) of the Marcus-Brundage tasks. Despite all the hype, agents didn’t turn out to be reliable.”

7. E7 No AGI by 2025 — Gary Marcus Prediction Scorecard · 1 January 2025

   “1 January 2025 · Annual predictions post on his Substack, Marcus on AI ("High confidence" list)”

8. E8 No AGI by 2025 — Gary Marcus Prediction Scorecard · 1 January 2025

   “No system recognized as AGI appeared in 2025; by late 2025 even leading figures such as Ilya Sutskever said scaling had stalled and new research was needed, putting human-level learning years away.”

9. E9 The People Building AI Are the Worst at Predicting It Whizi · 22 August 2026

   “Gary Marcus, January 2025 | 25 dated predictions, four graded here | All four correct, and none of the year's successes on the list | A-”

10. E10 Gary Marcus's AI Predictions: 19 tracked | Forecast Fools Forecast Fools · 28 July 2025

    “Forecast Fools tracks 19 rankable AI predictions from Gary Marcus, covering AGI, Markets, Regulation, Safety. 14 have resolved so far: 13 correct, 0 wrong, and 1 partial.”

E3 confirms that Marcus published a list of predictions for 2025 on 1 January 2025, and E8 refers to a "High confidence" list in that post, but none of the evidence lists all seventeen high-confidence items. Independent sources check only a few of them. LessWrong (E1/E5) found the Marcus-Brundage tasks prediction correct, Prediction Scorecard (E9) found the no-AGI prediction accurate, and Whizi (E10) graded four of the 25 predictions and found all four correct. Without an independent check of most of the seventeen, the evidence cannot show whether sixteen of them proved correct.