“Many leading figures in the field have acknowledged that we may have reached a period of diminishing returns of pure LLM scaling, much as I anticipated in 2022.”
Our reading, the sentence that was checked:
Many leading figures in the field have acknowledged that pure LLM scaling may have reached a period of diminishing returns, much as Marcus anticipated in 2022.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem, paragraph 16.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-sam-altman-thinks-that-agi-is-basically-d7eb37-a1
We recorded 16 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 2025-01-06. 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)
Conditions
- many leading figures in the field (scope, stated by the speaker)
- may have reached (qualifier, stated by the speaker)
From the speaker
- Many leading figures in the field have acknowledged
- may have reached
- a period of diminishing returns
- pure LLM scaling
- much as I anticipated in 2022
Our interpretation
- This is a claim about public acknowledgments made by publication, not a forecast that scaling must cease to produce improvements.
- Read 'pure LLM scaling' as increasing pretraining data, model size and compute without relying on distinct methodological advances or additional inference-time reasoning.
- Read 'many leading figures' as several prominent researchers or leaders of major AI organizations, without implying a majority or consensus. Marcus supplies no numerical count.
- The historical period begins in 2022 because the sentence explicitly includes his earlier warning and ends on the post date.
- Marcus's 2022 writing can establish the existence and wording of his warning. It cannot establish that the warning was technically correct.
Counts as supported if: Dated records show several prominent AI figures publicly acknowledging possible diminishing gains from pretraining scale alone before publication, and a 2022 record shows Marcus making a materially similar warning.
Counts as contradicted if: Reliable records establish that the attributed acknowledgments occurred only after publication or materially misstate those figures' positions, or that the identified 2022 statement did not make the claimed warning.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
By publication, Sutskever and Hassabis had publicly described reduced gains from scaling, while Pichai had explained that further progress required breakthroughs beyond scaling alone. Marcus's March 2022 article explicitly raised possible diminishing returns. These records support the sentence under the ordinary reading of 'many' as several prominent figures. The result depends on that unquantified reading and on counting Pichai's qualified acknowledgment. It does not establish consensus, prove a technical scaling limit, or validate every part of Marcus's 2022 argument.
Limitations
- The quantity 'many' remains interpretive.
- Pichai's acknowledgment is less explicit than those of Sutskever and Hassabis.
- News reporting documents the statements; an accessible original transcript was not obtained for Sutskever.
- The earlier Marcus article establishes prior wording only.
- Prominent contemporaries remained optimistic about continued progress.
Evidence (5 items)
Supports the claim: TIME,
performance gains from bigger models have plateaued.
Why it matters here: This contemporary report identifies Sutskever and explicitly describes his acknowledgment of a plateau. It is reputable secondary reporting of his position; the underlying Reuters interview was inaccessible through read_page.
- Publisher
- TIME
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: The New York Times, syndicated by The Indian Express,
no longer getting the same progress.
Why it matters here: The report attributes this statement to Google DeepMind leader Demis Hassabis and explicitly describes his view that returns from adding training data were diminishing. It also preserves his qualification that existing techniques could still deliver improvements.
- Publisher
- The New York Times, syndicated by The Indian Express
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: Semafor,
it’s not just a question of scaling alone
Why it matters here: In an original interview conducted Wednesday before publication, Google CEO Sundar Pichai explains that further progress requires breakthroughs beyond scaling alone. This supports the qualified reading of Marcus's sentence, although Pichai remained optimistic and did not declare that AI progress had stopped.
- Publisher
- Semafor
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: Nautilus,
perhaps already approaching a point of diminishing returns.
Why it matters here: The dated article directly establishes that Marcus voiced a similar warning in 2022, in a discussion that includes GPT-3-like models. It is used only as a record of his statement, not as evidence that his technical assessment was right.
- Publisher
- Nautilus
- Kind of source
- Secondary summary
- Describes events of
- Public from
Context only: The New York Times, syndicated by The Indian Express,
Not everyone
Why it matters here: The report also identifies Sam Altman, Dario Amodei and Jensen Huang as optimistic about continued progress. This limits any reading of Marcus's statement as an industry consensus, but does not negate the existence of the cited acknowledgments.
- Publisher
- The New York Times, syndicated by The Indian Express
- Kind of source
- News report
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Why I don’t share Sam Altman’s confidence that AGI is basically a solved problem
Many leading figures in the field have acknowledged that we may have reached a period of diminishing returns of pure LLM scaling, much as I anticipated in 2022.
- Where
- paragraph 16
- Published
- 2025-01-06
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-sam-altman-thinks-that-agi-is-basically-d7eb37-a1: 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-sam-altman-thinks-that-agi-is-basically-d7eb37-a1.