“Skeptical of this, I offered Musk a $1 million bet that AGI superior to the smartest individual humans would not arrive by the end of 2025, reprinted here for those who missed it.”
Our reading, the sentence that was checked:
AGI superior to the smartest individual humans will not arrive by the end of 2025.
Assessment: Supported Deadline passed on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Superhuman AGI is not nigh, paragraph 1.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-superhuman-agi-is-not-nigh-d7147d-a2
We recorded 14 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 2024-04-12. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- “by the end of 2025”
- Deadline state when assessed
- Deadline passed
- Deadline state on 2026-10-29
- Deadline passed (the release’s reference date)
From the speaker
- AGI superior to the smartest individual humans
- would not arrive by the end of 2025
- any intellectual work that any human can do
- at least one piece of intellectual work that the very best humans can do that AGI can’t
Our interpretation
- Use Marcus's explicit explanation: superiority must extend across human intellectual work, rather than mean a higher average score on selected tasks.
- Assess demonstrated practical capability using public evaluations of systems available by the deadline. Benchmark failures are evidence of capability gaps, not mathematical proof of impossibility.
- The wager amount is background, not a claim about company finances.
Counts as supported if: Evaluations covering the late-2025 frontier identify intellectual tasks that humans can solve but the leading AI systems still cannot reliably perform, with no demonstrated system closing that gap by the deadline.
Counts as contradicted if: A system available by December 31, 2025 demonstrates general intellectual capability meeting or exceeding the best humans across tasks, including independently tested unfamiliar tasks.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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
- Deadline passed
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The January 15, 2026 retrospective reports that every ARC-AGI-2 task was solved by at least two independent non-expert human testers [e1]. On December 5, 2025, ARC Prize's top verified refinement system scored 54% [e2]. On December 23, Poetiq reported up to 75% using GPT-5.2 X-High on the public evaluation set, a different split [e3]. These late-2025 results document remaining failures on human-solvable tasks under tested conditions. They support the forecast under the contract's public-capability standard and do not demonstrate superiority across human intellectual work by December 31, 2025 [e1, e2, e3].
Limitations
- The evidence does not inventory undisclosed systems or every system and inference configuration available by the deadline.
- Benchmark failures show gaps under tested conditions, not impossibility under every time, compute, or reliability allowance.
- Poetiq's December 23 result is developer-reported and uses a different evaluation split from the verified December 5 result.
- The assessment uses the claim's all-intellectual-work threshold; these benchmarks do not establish a general definition of AGI.
Evidence (3 items)
Supports the claim: ARC Prize Foundation, via arXiv,
hard for AI systems
Why it matters here: The retrospective report says every ARC-AGI-2 task was solved by at least two independent non-expert human testers. Its analysis identifies continuing generalization limitations in 2025 AI systems.
- Publisher
- ARC Prize Foundation, via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: ARC Prize Foundation,
scores 54%
Why it matters here: The organizer's leading verified result still left substantial failures on human-solvable reasoning tasks. This supports a remaining capability gap, although it is an early-December snapshot.
- Where
- Industry Progress
- Publisher
- ARC Prize Foundation
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: Poetiq,
~15 percentage points
Why it matters here: Poetiq reported up to 75% on the public evaluation set using GPT-5.2. This is material evidence of further progress, but uses a different evaluation split and does not demonstrate universal human-level task coverage.
- Publisher
- Poetiq
- 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, Superhuman AGI is not nigh
Skeptical of this, I offered Musk a $1 million bet that AGI superior to the smartest individual humans would not arrive by the end of 2025, reprinted here for those who missed it.
- Where
- paragraph 1
- Published
- 2024-04-12
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-superhuman-agi-is-not-nigh-d7147d-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-superhuman-agi-is-not-nigh-d7147d-a2.