“But one by one, almost every major thinker in AI has come around to the critique of LLMs that I began presenting in 2019.”
Our reading:
Almost every major AI thinker has come around to Marcus's critique of LLMs, first presented in 2019.
Assessment: Unresolved
What happened
Was it true for the period described?
Outcome: Unresolved
We checked, and the evidence so far doesn’t settle it.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
On March 21, 2023, Bengio and Hu argued that further scaling was unlikely to resolve current systems' reliability problems, showing overlap with the critique [e1]. In July 2025, Hassabis described roughly even chances that new breakthroughs would be needed or that scaling existing techniques would suffice [e2]. The International AI Safety Report, published January 29, 2025, recorded disagreement about scaling's ability to resolve remaining limitations [e4]. These sources establish neither near-universal adoption by September 26, 2025 nor substantial clear rejection of the specified critique, leaving the claim unresolved [e1] [e2] [e4].
Caveats
- The evidence provides no representative population or consistent measure of agreement among leading researchers.
- It does not establish convergence over time against a defined 2019 critique.
- Bengio and Hu's reliability argument does not establish agreement with the full proposition about general intelligence.
- Hassabis discusses scaling beyond pure language prediction.
- The safety report reflects an earlier evidence period and does not measure agreement in September 2025.
- Hinton's statements about understanding do not establish whether he considers scaling alone sufficient for general intelligence.
Evidence (4 items)
Supports the claim: Yoshua Bengio, co-written with Edward J. Hu,
further scaling is unlikely
Why it matters: This supports substantive overlap with the critique for one prominent researcher. It does not establish near-universality, a prior opposing position, or influence from Marcus.
- Publisher
- Yoshua Bengio, co-written with Edward J. Hu
- Kind of source
- Secondary summary
- About events on
- Published
Context only: Lex Fridman Podcast,
more room
Why it matters: Hassabis subsequently describes the need for new breakthroughs versus scaling existing techniques as roughly an even possibility. That is not a clear adoption of scaling's insufficiency, although his discussion includes more than pure language-model pretraining.
- Publisher
- Lex Fridman Podcast
- Kind of source
- News report
- About events on
- Published
Context only: University of Toronto,
understanding
Why it matters: The host's report also describes Hinton explicitly arguing that LLMs understand language. This cautions against treating leading researchers as sharing an undifferentiated critique, but does not establish his position on whether pure scaling alone achieves general intelligence.
- Publisher
- University of Toronto
- Kind of source
- News report
- About events on
- Published
Context only: International AI Safety Report,
Experts variously support
Why it matters: The report expressly describes disagreement about whether scaling can overcome remaining limitations. Its broader subject and earlier evidence cutoff prevent treating it as a direct census of agreement with Marcus in September 2025.
- Publisher
- International AI Safety Report
- Kind of source
- Secondary summary
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- Read the central shared critique as the view that scaling language prediction alone is insufficient for general intelligence.
- Acknowledging present weaknesses or proposing additional research does not by itself establish adoption of this critique.
- 'Come around' requires evidence of convergence over time, not merely a similar isolated statement.
- Do not substitute an arbitrary list of six researchers or a numerical percentage for the unnamed population of major AI thinkers.
- The claim describes developments from 2019 through the post date.
Right if: A broadly representative record of leading researchers shows near-universal adoption of substantially the same critique by September 26, 2025, with dated evidence supporting the asserted convergence and 2019 baseline.
Wrong if: A substantial body of leading researchers clearly continues to reject the core critique, or the asserted chronology is demonstrably wrong. One dissenter alone does not refute 'almost every'.
We wrote this test after the outcome was known.
What would settle it
A defined 2019 critique and representative, dated records of leading researchers' positions through September 26, 2025, showing near-universal adoption or substantial clear rejection and documenting any claimed changes of mind.
We found 6 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Game over for pure LLMs. Even Turing Award Winner Rich Sutton has gotten off the bus., paragraph 17.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2025-09-26.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Game over for pure LLMs. Even Turing Award Winner Rich Sutton has gotten off the bus.
But one by one, almost every major thinker in AI has come around to the critique of LLMs that I began presenting in 2019.
- Where
- paragraph 17
- Published
- 2025-09-26
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.
Held backBy us
The final check failed: a cited source didn’t say what the write-up said.