“The truth is that scaling just isn’t producing the fruit that it once did.”
Our reading, the sentence that was checked:
Scaling is no longer producing the gains it once did.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, GPT 4.5 is no GPT-5, paragraph 10.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-gpt-45-is-no-gpt-5-9d42b9-a1
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 2025-02-26. 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
- isn’t producing the fruit that it once did
Our interpretation
- Read 'fruit' as capability gains from increasing pretraining compute, model size, and training data, consistent with the surrounding discussion of pure scaling.
- The comparison concerns recent model generations against earlier generations, rather than merely whether models remain imperfect.
- Use January 1 through February 26, 2025 as the latest prepublication reporting window for the ongoing state described.
- No numerical decline threshold is added. Support requires a reasonably broad, comparable pattern of smaller capability gains for comparable increases in scale.
Counts as supported if: Comparable measurements across recent and earlier model generations show a broad decline in capability gains from equivalent increases in pretraining scale.
Counts as contradicted if: Comparable measurements show that recent pretraining scale increases continued to produce capability gains similar to or greater than earlier increases across a representative range of tasks.
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
Contemporaneous reporting supports concern about pretraining returns, and a primary experiment shows saturation on some tasks. The same experiment finds gains elsewhere. These records do not establish a broad decline relative to earlier generations using comparable resources and measurements. Remaining errors or disappointing releases alone cannot settle that comparison.
Limitations
- The available reporting is more specific to particular projects than the claim.
- The primary experiment concerns vision-language models and task-dependent results.
- Different reasonable measures of capability gains could change the assessment.
Evidence (2 items)
Context only: Fortune,
nearing its limits.
Why it matters here: Reports OpenAI-specific limits, without comparable measurements.
- Publisher
- Fortune
- Kind of source
- News report
- Describes events of
- Public from
Context only: Google DeepMind,
gains in low-resource languages
Why it matters here: The study reports saturation on several conventional benchmarks but continued gains on cultural-diversity and low-resource-language tasks. Its narrower vision-language setting does not settle the claimed broad historical change.
- Publisher
- Google DeepMind
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A documented comparison of earlier and early-2025 model generations using consistent capability measures and comparable increases in pretraining resources, with other major training changes accounted for.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, GPT 4.5 is no GPT-5
The truth is that scaling just isn’t producing the fruit that it once did.
- Where
- paragraph 10
- Published
- 2025-02-26
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-gpt-45-is-no-gpt-5-9d42b9-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.