“(vi) large language models would not get us to AGI and that we needed a new paradigm.”
Our reading:
Large language models will not get us to AGI; a new paradigm is needed.
Assessment: Unresolved
What happened
Did the forecast come true?
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
Marcus makes a forecast without a deadline. The reviewed evidence includes substantial capability gains and public AGI declarations, but no decisive event meeting both the capability and architectural criteria. ARC Prize reports strong performance while explicitly declining to call it AGI. The AGI Society assessment likewise finds the published evidence insufficient, and October robotics research reports both advances and limitations. None establishes that LLMs cannot eventually deliver AGI. Achieving AGI through another paradigm would also leave the claimed necessity unproved.
Caveats
- The meanings of AGI and new paradigm remain material interpretation choices.
- The AGI Society assessment is an evidence review, not a comprehensive experimental test.
- The robotics research is a preprint with a narrower scope than AGI.
- Failure to find decisive evidence is not proof that the forecast is correct.
Evidence (3 items)
Context only: ARC Prize,
It used fewer actions than the median tested human on 96% of levels.
Why it matters: This is material evidence of progress in learning unfamiliar environments. The same report explicitly declines to identify Astra as AGI and explains that the benchmark has bounded, deterministic tasks.
- Publisher
- ARC Prize
- Kind of source
- Independent measurement
- About events on
- Published
Context only: AGI Society,
Our current evaluation is that the published evidence does not yet support concluding that GPT-6 Astra represents the achievement Artificial General Intelligence.
Why it matters: The assessment examines public declarations that AGI has arrived and finds the published evidence insufficient. This challenges treating those declarations as decisive, but does not establish that LLMs can never reach AGI.
- Publisher
- AGI Society
- Kind of source
- Secondary summary
- About events on
- Published
Context only: arXiv; SenseNova and Joint Research Team,
Taken together, our results suggest that GPT-6-Astra has largely crossed the threshold of embodied competence, but not yet that of practical embodied intelligence.
Why it matters: The researchers report substantial embodied capabilities alongside practical limitations. Neither result establishes general intelligence or a permanent limitation of the LLM paradigm; embodiment is not imposed here as an additional AGI requirement.
- Publisher
- arXiv; SenseNova and Joint Research Team
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
How we judged it
- Kind of claim
- Forecast
- Deadline
- None given. We don’t add one.
Where we had to interpret
- The plain reading is that extending the LLM approach cannot deliver broadly human-level general intelligence; a fundamental change in approach is necessary.
- AGI means broadly human-level ability to learn and perform across diverse intellectual tasks, including unfamiliar tasks. Marcus supplies no specific test or numerical threshold.
- The relevant LLM approach is the approach discussed in the 2022 argument and reaffirmed in this 2023 post. Scaling, prompting, or fine-tuning alone would not constitute a new paradigm. Whether added reasoning mechanisms, memory, tools, or architectural changes constitute a new paradigm requires explicit technical assessment.
- An LLM-based system is a potential counterexample only if its general capabilities do not depend on a fundamental change beyond that approach. Merely containing an LLM is insufficient.
- This is an indefinite forecast, not an observation that AGI had not yet arrived. Success through another paradigm would not establish that the change was necessary.
Right if: Compelling, independently scrutinized evidence establishes an inherent limitation of the relevant LLM approach that prevents a capability necessary for AGI, and establishes why overcoming it requires a fundamental change of paradigm. Failure of existing models, or AGI achieved through another approach, is insufficient by itself.
Wrong if: A publicly demonstrated system achieves broadly human-level general intelligence, with credible independent evaluations across diverse and unfamiliar tasks, and sufficient technical evidence establishes that it achieves this within the relevant LLM paradigm without a necessary fundamental change of approach.
We wrote this test after the outcome was known.
What would settle it
An independently validated demonstration of broadly human-level general intelligence, with technical evidence that it remains within the relevant LLM paradigm, would contradict the claim. Establishing support requires evidence that an essential AGI capability cannot be obtained within that paradigm and therefore requires a fundamental change; another approach succeeding first would not suffice.
We found 4 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Has Sam Altman gone full Gary Marcus?, paragraph 2.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-29. The post may have been edited after 2023-11-16.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Has Sam Altman gone full Gary Marcus?
(vi) large language models would not get us to AGI and that we needed a new paradigm.
- Where
- paragraph 2
- Published
- 2023-11-16
- 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
A reviewer said our test wasn’t fair to what was actually said.