“For nearly three years I have been saying that the pure scaling of LLMs – adding more data and more parameters – would eventually run out, and that it would fail to solve hallucinations and boneheaded errors and that scaling laws were merely empirical generalizations, rather than physical laws.”
Our reading, the sentence that was checked:
Marcus restates his long-held prediction that pure scaling of LLMs (adding more data and more parameters) would eventually run out and would fail to solve hallucinations and boneheaded errors, and that scaling laws are merely empirical generalizations rather than physical laws.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Breaking: OpenAI's efforts at pure scaling have hit a wall., paragraph 3.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-breaking-openais-efforts-at-pure-4dee04-a1
We recorded 9 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 2025-02-13. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
Conditions
- restated as something he has been saying for nearly three years (qualifier, stated by the speaker)
- pure scaling meaning adding more data and more parameters (scope, stated by the speaker)
- the scaling-laws point is a description, not a prediction (qualifier, stated by the speaker)
From the speaker
- pure scaling of LLMs
- adding more data and more parameters
- would eventually run out
- fail to solve hallucinations and boneheaded errors
- scaling laws were merely empirical generalizations, rather than physical laws
Our interpretation
- Assess the restated technical prediction, not the separate assertion about how long Marcus had advocated it.
- Read 'run out' as reaching a limit on substantial general capability improvements from increasing data and parameters alone.
- Read 'solve' as reliably overcoming recurring hallucinations and basic errors across ordinary tasks, rather than guaranteeing perfection on every conceivable input.
- The description of scaling laws is a separate factual component. Establishing it does not settle the forecast.
Counts as supported if: Evidence establishes a general limit to capability gains from scaling data and parameters alone, with hallucinations and basic errors remaining unresolved. The scaling-laws component requires evidence that the laws are empirical relationships.
Counts as contradicted if: A reproducibly evaluated system developed through additional data and parameters alone reliably overcomes the specified error classes across broad tasks. Continued improvements alone do not disprove an eventual limit.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
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 deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The empirical nature of scaling laws is supported, and hallucinations remained present in the cited research. Neither fact establishes the forecast's eventual limit. Reported improvements also caution against declaring scaling exhausted, but do not disprove an eventual ceiling. No decisive event meets the early-resolution rule.
Limitations
- Available model comparisons change several factors simultaneously.
- Evidence of present errors cannot establish a permanent inability to solve them.
- The descriptive scaling-laws component does not resolve the bundled forecast.
Evidence (3 items)
Context only: Jared Kaplan and coauthors, arXiv,
empirical scaling laws
Why it matters here: This supports the descriptive component about empirical relationships. It does not establish where their usefulness ends or whether scaling can solve the named errors.
- Publisher
- Jared Kaplan and coauthors, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Adam Tauman Kalai and coauthors, arXiv,
persist even in state-of-the-art systems
Why it matters here: Persistent hallucinations are consistent with the prediction. Their persistence in existing systems does not establish that additional data and parameters can never overcome them.
- Publisher
- Adam Tauman Kalai and coauthors, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI,
pre-training and post-training
Why it matters here: OpenAI reports further improvements, which cautions against treating scaling as already exhausted. The disclosure includes other training changes, so it is not a controlled test of pure scaling or a disproof of an eventual limit.
- Where
- Opening paragraph
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Reproducible evidence establishing a general scaling-only ceiling that leaves the specified failures unresolved, or a documented scaling-only system that reliably overcomes those failures across broad tasks.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Breaking: OpenAI's efforts at pure scaling have hit a wall.
For nearly three years I have been saying that the pure scaling of LLMs – adding more data and more parameters – would eventually run out, and that it would fail to solve hallucinations and boneheaded errors and that scaling laws were merely empirical generalizations, rather than physical laws.
- Where
- paragraph 3
- Published
- 2025-02-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-breaking-openais-efforts-at-pure-4dee04-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-breaking-openais-efforts-at-pure-4dee04-a1.