“It also drives their greediness with data, and even the ethical choices their developers have been making.”
Our reading, the sentence that was checked:
Neural networks' trouble generalizing beyond their training space, at least in the configurations dominant over the last three decades, also drives their greediness with data and even the ethical choices their developers have been making.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Breaking news: Scaling will never get us to AGI, paragraph 1.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-breaking-news-scaling-will-never-84c11a-a2
We recorded 6 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 2024-04-08. 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)
Conditions
- at least in the configurations that have been dominant over the last three decades (scope, stated by the speaker)
- ethical choices their developers have been making (scope, stated by the speaker)
From the speaker
- trouble generalizing beyond the multidimensional space that surrounds their training examples
- drives their greediness with data
- the ethical choices their developers have been making
- configurations that have been dominant over the last three decades
Our interpretation
- I read greediness as needing many training examples to achieve useful performance.
- I read ethical choices as consequential decisions about data collection, consent, licensing and privacy.
- The claim alleges a causal chain from weak generalization to data demand and then to developer decisions. Showing large datasets and controversial conduct separately is insufficient.
- The assessment concerns configurations and developer conduct by the post date.
Counts as supported if: Technical evidence links generalization limitations to increased data needs, and documented developer decisions connect those needs to consequential collection or licensing choices.
Counts as contradicted if: Controlled technical evidence and documented decisions show that the alleged generalization limitation did not materially drive the data demand or the developer choices.
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
Lake and Baroni's 2018 study found systematic compositional generalization failures in tested recurrent networks and tentatively linked lack of systematicity to training-data needs [e1]. Reporting available on April 8, 2024 described Meta's 2023 discussions of collecting copyrighted data despite potential lawsuits and the time required for licensing [e2]. The study does not connect its technical mechanism to those decisions, and the reporting does not isolate generalization limitations as their cause [e1, e2]. The full causal chain is neither established nor refuted [e1, e2].
Limitations
- The technical evidence concerns particular recurrent networks and tasks, and its link to data demand is tentative [e1].
- The corporate discussions are documented through reporting based on internal recordings; the underlying records were not independently verified in this packet [e2].
- The evidence does not distinguish data needed for unfamiliar facts from data needed to compensate for weak generalization in the reported decisions [e1, e2].
Evidence (2 items)
Context only: ICML; arXiv,
RNNs fail spectacularly.
Why it matters here: The study demonstrates a generalization weakness and tentatively connects lack of systematicity to data needs. It does not establish that this mechanism caused particular companies' collection decisions.
- Where
- Abstract
- Publisher
- ICML; arXiv
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: The New York Times, republished by The Indian Express,
even if that meant facing lawsuits.
Why it matters here: Reporting based on meeting recordings connects the pursuit of training data to consequential licensing decisions. It does not isolate weak generalization as the cause; competition, cost and licensing speed also appear in the account.
- Publisher
- The New York Times, republished by The Indian Express
- Kind of source
- News report
- Describes events of
- Public from
What would settle it
Controlled studies separating factual coverage from generalization-related data needs, combined with decision records linking those needs to collection, consent or licensing choices within the period ending April 8, 2024. Evidence establishing that mechanism would support the claim; controlled evidence and records showing that the alleged limitation did not materially drive demand or decisions would contradict it.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Breaking news: Scaling will never get us to AGI
It also drives their greediness with data, and even the ethical choices their developers have been making.
- Where
- paragraph 1
- Published
- 2024-04-08
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-breaking-news-scaling-will-never-84c11a-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-breaking-news-scaling-will-never-84c11a-a2.