“But a lot of things that a bunch of us worried about 30 years ago remain true.”
Our reading, the sentence that was checked:
Many concerns raised about neural networks 30 years ago still hold true.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Happy Groundhog Day, The AI Edition, paragraph 3.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-happy-groundhog-day-the-ai-edition-889f4f-a2
We recorded 12 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 2023-02-03. 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
- a lot of things
- worried about 30 years ago
- remain true
- compositionality
- abstraction
- plain old silly errors
Our interpretation
- The claim concerns recurring problems in combining familiar parts, transferring rules to unfamiliar cases, and producing correct answers. It does not assert that every historical criticism remains valid.
- For the ongoing state, use the latest completed research year, 2022, plus the portion of 2023 preceding the post. Research evaluations have no standard quarterly reporting period.
- The older criticism supplies historical context; contemporary evaluations must establish persistence.
- Improved performance is compatible with unresolved problems.
Counts as supported if: Independent contemporary evaluations show recurring failures across several of the named concerns, and the historical record establishes that these concerns predate the current revival.
Counts as contradicted if: Broad contemporary evaluations establish reliable performance on the named abilities, leaving only isolated examples that do not sustain the claim of recurring problems.
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
Fodor and Pylyshyn's March 1988 paper documents concerns about compositional and systematic representations [e1]. In December 2022, controlled T5 tests found failures to transfer familiar linguistic components into new contexts [e2]. The June 2022 BIG-bench evaluation found substantial remaining weaknesses across model families and tasks [e3]. Together, these findings support persistence of concerns about composition, transfer and answer accuracy [e1–e3]. The 2022 studies also reported improvements from scale and prompting, so persistence does not imply an absence of progress [e3, e4].
Limitations
- The evaluations do not cover every historical criticism or every neural architecture [e2–e4].
- BIG-bench's reported aggregate scores do not establish an everyday-use error rate [e3].
- The contemporary evaluations do not establish the historical paper's broader architectural conclusions [e1–e4].
Evidence (4 items)
Supports the claim: arXiv; Najoung Kim, Tal Linzen and Paul Smolensky,
contextual difficulty
Why it matters here: Controlled changes to COGS exposed failures to transfer familiar linguistic components into new contexts in T5. This directly supports persistence of compositional generalization problems, rather than merely unfamiliar-word recognition problems.
- Publisher
- arXiv; Najoung Kim, Tal Linzen and Paul Smolensky
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: arXiv; BIG-bench collaboration,
perform poorly overall
Why it matters here: The evaluation finds persistent errors across a broad collection of reasoning and knowledge tasks and multiple model families. The same section reports improvement with model scale and examples, so these results support remaining weaknesses rather than no progress.
- Where
- Section 3.1
- Publisher
- arXiv; BIG-bench collaboration
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Cognition; Jerry Fodor and Zenon Pylyshyn,
combinatorial structure
Why it matters here: The original paper establishes the historical concern about compositional and systematic representations. Its March publication is recorded conservatively at month end; it does not establish the performance of modern models.
- Where
- Abstract
- Publisher
- Cognition; Jerry Fodor and Zenon Pylyshyn
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: arXiv; Ofir Press and colleagues,
further improves
Why it matters here: The study shows that prompting can improve the composition of intermediate answers. This limits a stronger interpretation that the old problems had seen no meaningful progress, but does not establish their general solution.
- Publisher
- arXiv; Ofir Press and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Happy Groundhog Day, The AI Edition
But a lot of things that a bunch of us worried about 30 years ago remain true.
- Where
- paragraph 3
- Published
- 2023-02-03
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-happy-groundhog-day-the-ai-edition-889f4f-a2: 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-happy-groundhog-day-the-ai-edition-889f4f-a2.