“Deep learning is pretty decent at, for example, recognizing objects, but not nearly as good at planning, or reading, or language comprehension.”
Our reading, the sentence that was checked:
Deep learning is decent at object recognition but much worse at planning, reading and language comprehension.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Dear Elon Musk, here are five things you might want to consider about AGI, paragraph 20.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-dear-elon-musk-here-are-five-things-5dc69d-a1
We recorded 21 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 2022-05-31. 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
- Deep learning
- pretty decent at
- recognizing objects
- not nearly as good at planning, or reading, or language comprehension
Our interpretation
- The ordinary reading compares the maturity and reliability of deep learning across the named abilities, rather than comparing unrelated benchmark percentages.
- Reading is understood as understanding written material, given its pairing with language comprehension.
- Human-relative task reliability is a useful comparison standard, but Marcus supplies neither this measure nor a required size of gap.
- The observation concerns the current 2022 research record through May 31; academic research has no fixed reporting quarter.
Counts as supported if: Representative, reasonably comparable evaluations establish useful object recognition and substantially weaker reliability in planning, reading, and language comprehension.
Counts as contradicted if: Representative, reasonably comparable evaluations establish that the claimed broad disadvantage in the other abilities does not hold.
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 evidence establishes strong object recognition and real planning limitations. It also establishes strong language benchmark performance before the post. These findings do not settle the comparative claim: the tasks, metrics, and evaluation conditions differ. Calling the whole statement supported or contradicted would depend too heavily on which tasks the evaluator selected.
Limitations
- Benchmark success does not establish broad real-world comprehension.
- A planning failure does not by itself establish that planning was substantially worse than recognition.
- The evidence supports individual observations without resolving the claimed ranking.
Evidence (3 items)
Context only: arXiv; Jiahui Yu and colleagues,
91.0% on ImageNet
Why it matters here: This supports the component claim that deep learning was effective at object recognition. The different task scores do not establish a common scale for comparing recognition with planning or comprehension.
- Publisher
- arXiv; Jiahui Yu and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Wenlong Huang and colleagues,
admissible actions
Why it matters here: The study documents a practical planning limitation while also finding useful planning ability and improvements from translating plans into admissible actions. It does not compare planning reliability with object recognition.
- Where
- Abstract
- Publisher
- arXiv; Wenlong Huang and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Aakanksha Chowdhery and colleagues,
language understanding
Why it matters here: PaLM reports strong language results, including performance above the average human result on its BIG-bench evaluation. This challenges a blanket characterization of weak language ability but does not supply the required comparison with object recognition.
- Where
- Abstract
- Publisher
- arXiv; Aakanksha Chowdhery and colleagues
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A representative evaluation of systems available by May 2022 using a defensible common standard across all four abilities, with an explicit treatment of task difficulty and generalization.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Dear Elon Musk, here are five things you might want to consider about AGI
Deep learning is pretty decent at, for example, recognizing objects, but not nearly as good at planning, or reading, or language comprehension.
- Where
- paragraph 20
- Published
- 2022-05-31
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-dear-elon-musk-here-are-five-things-5dc69d-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.
About assessment marcus-dear-elon-musk-here-are-five-things-5dc69d-a1.