“We in AI have a long long way to go. Data and compute alone won’t be enough.”
Our reading, the sentence that was checked:
AI has a long way to go, and data and compute alone will not be enough.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Sub-Optimal, paragraph 59.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-sub-optimal-0d9437-a1
We recorded 11 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 2022-10-01. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Causal claim
- 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
- We in AI have a long long way to go
- Data and compute alone won’t be enough
Our interpretation
- Enough refers to the surrounding goal of general intelligence and reliable common sense, especially in real-world robotics.
- Data and compute alone means increasing those resources while holding the underlying learning and inference methods substantially fixed.
- Long long way to go supplies no measurable distance or deadline and is treated as emphasis.
- The claim concerns sufficiency, not whether additional data and compute are useful.
Counts as supported if: Causal or theoretical evidence establishes that increasing data and compute within the existing approach is insufficient for the contextual capability target and that another essential ingredient is required.
Counts as contradicted if: A reproducibly evaluated system reaches the contextual goal through increases in data and compute alone, with adequate disclosure and controls excluding essential method changes.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Does the evidence support the cause?
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 records show useful gains from resource allocation, substantial recent benchmark performance and continuing generalization problems. These are compatible with several explanations. They establish neither that data and compute alone suffice for the contextual goal nor that another ingredient is necessarily required.
Limitations
- Benchmark performance covers only part of the capability claim.
- The distinction between resource scaling and method changes is underspecified.
Evidence (3 items)
Context only: ARC Prize Foundation, arXiv,
extreme bimodal performance
Why it matters here: The experiment illustrates why success on familiar tasks does not establish broad generalization and why surrounding procedures matter. It does not prove that additional data and compute could never overcome the observed limits.
- Publisher
- ARC Prize Foundation, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Jordan Hoffmann and coauthors, arXiv,
67.5%
Why it matters here: The study demonstrates meaningful gains from changing the allocation of model size and training data under a compute budget. It weighs against dismissing those resources, but benchmark improvement does not settle their ultimate sufficiency.
- Publisher
- Jordan Hoffmann and coauthors, arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: ARC Prize,
90.2%
Why it matters here: These recent results demonstrate substantial benchmark capability and must be considered alongside earlier limitations. They neither establish the full contextual goal nor identify data and compute alone as the cause.
- Publisher
- ARC Prize
- Kind of source
- Benchmark result
- Describes events of
- Public from
What would settle it
Controlled, independently reproduced evidence that resource increases alone achieve the contextual goal, or strong evidence establishing a relevant limitation that necessarily requires another ingredient.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Sub-Optimal
We in AI have a long long way to go. Data and compute alone won’t be enough.
- Where
- paragraph 59
- Published
- 2022-10-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-sub-optimal-0d9437-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.