Pundit ClaimsAI claims by Gary Marcus and Ed Zitron, with the evidence

Gary Marcus Observation First said

“Scaling has run out (like I said it would)”

Scaling is over, the bubble may be deflating, LLMs still can’t reason, and you can’t trust Sam, paragraph 12. The full statement is at the source.

Our reading, the sentence that was checked:

Scaling of LLMs has run out as of April 2025.

Assessment: Contradicted Evidence as of

Where and when Gary Marcus wrote it

Marcus on AI, Scaling is over, the bubble may be deflating, LLMs still can’t reason, and you can’t trust Sam, paragraph 12.

Said
Source observed
live page, full text, public
Evidence as of
the cutoff for this assessment
Assessment recorded
assessment marcus-scaling-is-over-the-bubble-may-be-d00cb4-a2

We recorded 8 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 2025-04-06. 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

  • Scaling has run out
  • Model and data size scaling are over
  • 30 trillion training tokens and 2 trillion parameters

Our interpretation

  • The post's opening discussion makes model and pretraining-data scaling the relevant subject.
  • The ordinary reading is that increasing pretraining scale no longer delivers meaningful capability gains at the frontier, rather than merely delivering diminishing returns.
  • A larger non-reasoning model need not outperform every smaller reasoning model for pretraining scaling to remain useful.
  • The target period is the early-2025 model generation through the post date. Unlike financial reporting, model evaluations have no fixed reporting quarter; this period choice is an interpretation.
  • Later analysis may be used only for results concerning models and scaling steps already present in the target period.

Counts as supported if: Comparable frontier evaluations show that increasing model, data, or pretraining-compute scale has ceased to produce meaningful capability gains.

Counts as contradicted if: Comparable evaluations tied to increased pretraining scale show meaningful capability gains during the target period.

Retrospective: these criteria were written after the outcome was public.

Current assessment

Was it true for the period described?

Outcome: Contradicted

The dated evidence goes against 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
No horizon
Assessment recorded
Policy version
2026-10-28.7

Rationale

On February 27, 2025, OpenAI reported that GPT-4.5's increased pretraining scale brought gains over GPT-4o, including 71.4% versus 53.6% on GPQA and 36.7% versus 9.3% on AIME 2024 [e1]. Epoch AI's September 16, 2025 analysis of that February model generation found its gains consistent with expected compute-scaling trends [e2]. These results contradict the reading that pretraining scaling had ceased to deliver meaningful capability gains by April 6, 2025 [e1, e2].

Limitations

  • The assessment interprets 'run out' as cessation of meaningful capability gains, rather than poor economic returns.
  • The comparisons do not isolate increased scale from concurrent architectural and training-method changes.
  • OpenAI's benchmark results are company-reported.
  • The evidence does not establish whether the gains justified the additional cost.

Evidence (2 items)

  • Goes against the claim: OpenAI,

    shows meaningful improvements

    Why it matters here: OpenAI ties the larger pretraining run to gains: its table reports GPQA scores of 71.4% versus 53.6% for GPT-4o and AIME 2024 scores of 36.7% versus 9.3%. These are company-reported comparisons; the release also acknowledges architectural and optimization changes.

    Publisher
    OpenAI
    Kind of source
    Company disclosure: the company describing itself, not an independent finding
    Describes events of
    Public from
  • Goes against the claim: Epoch AI,

    scaling roughly as expected.

    Why it matters here: The footnote compares GPT-4.5's gains over GPT-4o on GPQA and Math Level 5 with observed performance-versus-compute trends. This is later analysis of the February model generation, not evidence from a later model.

    Publisher
    Epoch AI
    Kind of source
    Independent measurement
    Describes events of
    Public from

Where it was said (1)

Said once in the sources this register covers.

  • : Marcus on AI, Scaling is over, the bubble may be deflating, LLMs still can’t reason, and you can’t trust Sam

    Scaling has run out (like I said it would)

    Published
    2025-04-06
    Source observed
    , live page
    Access
    full text, public

History

Assessment versions

Recorded events

  1. Assessment publishedBy this site

    First assessment published after both reviews and the final-text check passed.

    About assessment marcus-scaling-is-over-the-bubble-may-be-d00cb4-a2.