“Scaling has undeniably helped, but not solved any of the problems I pointed to above.”
Our reading, the sentence that was checked:
Scaling has helped but has not solved any of the fundamental deep-learning problems Marcus identified.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Two years later, deep learning is still faced with the same fundamental challenges, paragraph 20.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-two-years-later-deep-learning-is-c3a58d-a2
We recorded 15 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2024-03-10. 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 undeniably helped
- not solved any
- the problems I pointed to above
Our interpretation
- This describes accumulated results through the post, rather than predicting that scaling can never solve the problems.
- The context explicitly starts with the 2020 scaling work, so the target period is 2020 through March 10, 2024.
- Helped means measurable gains from increased model size, training data or compute.
- Solved means reliably overcoming a failure class in its intended applications, not achieving literally zero errors.
- The testable core includes hallucination, simple errors, toxicity and misinformation. The word any also reaches broader preceding concerns about understanding, trustworthy operation and general-purpose physical assistance; I do not silently discard that wider scope.
Counts as supported if: Scaling studies establish capability gains, while evidence covering every identified problem shows it remained materially unsolved through the post.
Counts as contradicted if: Evidence shows no meaningful gains attributable to scaling, or demonstrates that scaling reliably overcame at least one of the identified problem classes.
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
The 2021 Gopher study documents capability gains from increased parameters and training compute while holding the dataset and token count fixed [e1]. The GPT-4 report available on March 4, 2024 documents continuing hallucinations and reasoning errors alongside factuality improvements [e2]. These findings support scaling gains and persistent failures in the tested systems, but do not establish that every referenced problem remained materially unsolved through March 10, 2024 [e1, e2].
Limitations
- The evidence does not cover every referenced problem or comprehensively evaluate the strongest systems through March 10, 2024.
- The packet supplies no application-specific reliability thresholds for deciding whether each problem was solved.
- GPT-4's improvements do not isolate scaling from other training and safety interventions.
Evidence (2 items)
Supports the claim: DeepMind researchers, arXiv,
52.7% accuracy
Why it matters here: Comparisons within the Gopher family hold the dataset and token count fixed and show that increasing parameters and training compute improved many tasks. This supports the gains component but cannot establish that every broader problem remained unsolved in March 2024.
- Publisher
- DeepMind researchers, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: OpenAI, arXiv,
it “hallucinates” facts
Why it matters here: The report documents continuing factual and reasoning failures, and says safeguards can still be bypassed. It also reports substantial safety gains, including toxic-generation rates of 0.73% versus 6.48% for GPT-3.5 on its stated benchmark; these combined interventions do not isolate scaling.
- Publisher
- OpenAI, arXiv
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Identify all referenced problems and define application-specific reliability thresholds. Provide evaluations covering 2020 through March 10, 2024 that establish whether every problem remained materially unsolved or scaling reliably overcame at least one, separating scaling from other interventions.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Two years later, deep learning is still faced with the same fundamental challenges
Scaling has undeniably helped, but not solved any of the problems I pointed to above.
- Where
- paragraph 20
- Published
- 2024-03-10
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-two-years-later-deep-learning-is-c3a58d-a2: 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-two-years-later-deep-learning-is-c3a58d-a2.