“Neither one of us believes for a minute that large language models are remotely close to artificial general intelligence, and what’s more, we mostly agree about what’s missing: semantics, reasoning, common sense, theory of mind and so on.”
Our reading, the sentence that was checked:
Large language models are not remotely close to AGI; they lack semantics, reasoning, common sense and theory of mind.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, AGI will not happen in your lifetime. Or will it?, paragraph 8.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-agi-will-not-happen-in-your-lifetime-49f087-a1
We recorded 28 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 2023-01-22. 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
- large language models
- remotely close to artificial general intelligence
- what’s missing: semantics, reasoning, common sense, theory of mind
Our interpretation
- Assess the technical assertion, rather than whether Marcus accurately describes his and Booch's beliefs.
- The ordinary contextual reading is that LLMs lack reliable, general competence in the listed abilities. It need not mean that they cannot solve any reasoning or social-understanding task.
- Use January 2023 through the publication date as the current research snapshot. The post specifies no reporting period.
- Do not substitute later models for models available when the statement was made.
Counts as supported if: Evaluations of contemporary LLMs establish substantial, general deficiencies across the named abilities and connect those deficiencies to a defensible measure of distance from human-level general intelligence.
Counts as contradicted if: Contemporary LLMs demonstrate reliable, transferable competence across the named abilities and broad human-level generality. Isolated benchmark successes challenge a literal absence reading but do not alone contradict the contextual claim.
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 research documents both substantial weaknesses and real reasoning successes before the post. It therefore supports concern about reliability while challenging a literal claim that the abilities were entirely absent. Neither finding establishes how close January 2023 LLMs were to AGI. A supported, contradicted or mixed verdict would depend too heavily on supplying a threshold Marcus did not give.
Limitations
- Historical benchmark results are context, not a complete January 2023 assessment.
- No later model result is used to overturn this historical observation.
Evidence (3 items)
Context only: Srivastava et al., arXiv,
We find the strong dependence on prompting striking.
Why it matters here: The study documents substantial changes in causal-task performance when questions are reformulated. These earlier experiments support concern about robustness, but do not measure the January 2023 field's distance from AGI.
- Publisher
- Srivastava et al., arXiv
- Kind of source
- Benchmark result
- Describes events of
- Public from
Context only: Google Research,
improves performance on various types of commonsense reasoning tasks.
Why it matters here: Reasoning and commonsense benchmark successes existed before the post. This challenges a claim of complete absence, while leaving reliability and broad generality unsettled.
- Publisher
- Google Research
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Context only: Sap et al., arXiv,
only 55% accuracy
Why it matters here: This earlier GPT-3 evaluation documents a substantial social-reasoning shortfall. It covers selected models and tasks, not every contemporary LLM or every named faculty.
- Publisher
- Sap et al., arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
What would settle it
A defensible AGI proximity criterion and comprehensive evaluations of the January 2023 model versions against it, including robust tests of all named abilities.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, AGI will not happen in your lifetime. Or will it?
Neither one of us believes for a minute that large language models are remotely close to artificial general intelligence, and what’s more, we mostly agree about what’s missing: semantics, reasoning, common sense, theory of mind and so on.
- Where
- paragraph 8
- Published
- 2023-01-22
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-agi-will-not-happen-in-your-lifetime-49f087-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-agi-will-not-happen-in-your-lifetime-49f087-a1.