“And it would be absurd to say that o3 can do all tasks that humans can do in 1.7 hours or that GPT-3 could do anything that a person can do in 15 seconds”
Our reading, the sentence that was checked:
o3 cannot do all tasks humans can do in 1.7 hours, and GPT-3 could not do everything a person can do in 15 seconds.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The latest AI scaling graph - and why it hardly makes sense, paragraph 17.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-latest-ai-scaling-graph-and-why-49a50a-a1
We recorded 5 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 2025-05-03. 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
- o3
- all tasks that humans can do in 1.7 hours
- GPT-3
- anything that a person can do in 15 seconds
Our interpretation
- Read anything in parallel with all tasks: GPT-3 lacks universal coverage, rather than being incapable of every short task.
- Assess cognitive or computer tasks relevant to the graph, excluding trivial counterexamples based on lacking a physical body.
- Read the durations as upper limits on human completion time.
- The period covers the historical GPT-3 capability comparison and the o3 capabilities available by the post. It does not include later successor models.
- Require evidence linking model failures to human completion times, rather than assuming that an apparently easy task satisfies the limit.
Counts as supported if: For each named model, credible evaluations demonstrate failures on tasks that humans successfully complete within its stated time limit.
Counts as contradicted if: Evidence establishes the asserted universal task coverage for the relevant model and time limit. High average benchmark scores alone would not do so.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
Primary evaluations show failures by both models, consistent with Marcus’s claim. But the retrieved evidence does not fully connect those failures to the specified human time limits, particularly GPT-3’s 15-second limit. Treating arithmetic as automatically satisfying that limit would substitute an assumption for the requested evidence. The compound claim therefore remains unresolved under the matched-timing criteria.
Limitations
- Unresolved reflects an evidence gap, not evidence that either model has universal capability.
- The cited evaluations test particular configurations.
- Aggregate benchmark failure and aggregate human speed are insufficient for the strict task-level comparison used here.
Evidence (2 items)
Context only: ARC Prize Foundation,
Despite recent gains, ARC-AGI-2 remains unsolved by the best version of o3, with scores below 3%.
Why it matters here: This establishes substantial failures by released o3 in the evaluated configurations. It does not itself establish the human duration of a matching failed task, and high-reasoning evaluation coverage was incomplete.
- Publisher
- ARC Prize Foundation
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: OpenAI researchers via arXiv,
GPT-3 also achieves 29.2% accuracy at 2 digit multiplication, an especially computationally intensive operation.
Why it matters here: The primary report establishes GPT-3 arithmetic failures. It supplies no matched human timing, so treating these as demonstrated 15-second counterexamples would add an unsupported assumption.
- Publisher
- OpenAI researchers via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A dated, reproducible counterexample for each model, paired with successful human completion within 102 minutes and 15 seconds respectively.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, The latest AI scaling graph - and why it hardly makes sense
And it would be absurd to say that o3 can do all tasks that humans can do in 1.7 hours or that GPT-3 could do anything that a person can do in 15 seconds
- Where
- paragraph 17
- Published
- 2025-05-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-latest-ai-scaling-graph-and-why-49a50a-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-the-latest-ai-scaling-graph-and-why-49a50a-a1.