“we used to actually have many different approaches to AI, and the fields settled on one, the large language model, which is the best prediction machine that we have. And it does that very well, but it doesn’t do some of the other things very well.”
Our reading:
The AI field once had many different approaches but settled on one, the large language model, which is the best prediction machine we have and predicts very well but doesn't do some of the other things very well.
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
The historical diversity component has documentary support. Current world-simulation work complicates an exclusive reading of convergence, but does not disprove LLM dominance. The reasoning study establishes narrower limitations in older models. The searches did not establish either October-wide convergence or the unspecified claim that LLMs are the best prediction machines, so the combined statement remains unresolved.
Caveats
- Dominance and exclusivity would produce different tests.
- Language prediction and prediction across all domains are different comparison classes.
- The company announcement is not an independent performance evaluation.
Evidence (3 items)
Context only: AAAI,
many different methodologies
Why it matters: The report supports the historical diversity component but cannot establish the October 2026 state. It identifies March 2025 as its publication month; available_at conservatively uses that month's final day.
- Publisher
- AAAI
- Kind of source
- Secondary summary
- About events on
- Published
Context only: Google,
interactive simulation.
Why it matters: Google documents an October world-simulation initiative, complicating an exclusive reading of settled on one. It does not measure the prevalence of LLMs or prove that this work is independent of language-model components.
- Publisher
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: arXiv,
significantly underperform
Why it matters: The reported evaluation shows weaknesses in interactive knowledge reasoning, but its leading tested model is GPT-5 and its other examples include earlier generations. It does not establish the October frontier's overall performance profile or comparative prediction leadership.
- Where
- Abstract
- Publisher
- arXiv
- Kind of source
- Preprint, not peer reviewed
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Conditions
- some of the other things (their words)
Where we had to interpret
- Read settled on one as LLMs becoming the prevailing approach to general-purpose AI, rather than the literal disappearance of every other method.
- In the surrounding discussion, prediction most naturally refers to language prediction. A claim about superiority across every prediction domain would be substantially broader.
- The surrounding examples identify reasoning and instruction following as the other abilities.
- Use October 2026 for the ongoing state and performance claims. Earlier diversity is a historical comparison with no specified starting date.
- Some of the other things is a scope qualification, not a conditional forecast antecedent.
Right if: Research or deployment records establish convergence on LLMs as the prevailing general-purpose approach; matched comparisons establish their claimed prediction leadership; contemporary evaluations establish significant weaknesses in the specified other abilities.
Wrong if: Matched evidence refutes a material component: no such convergence, superior prediction by another approach on the intended tasks, or consistently strong performance across the other abilities being described.
We wrote this test after the outcome was known.
What would settle it
Contemporary systematic records of research or deployment, a specified prediction comparison, and matched evaluations of prediction, reasoning, and instruction following.
We found 32 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Report from the Vatican, paragraph 22.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-11-09. The post may have been edited after 2026-10-31.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Report from the Vatican
we used to actually have many different approaches to AI, and the fields settled on one, the large language model, which is the best prediction machine that we have. And it does that very well, but it doesn’t do some of the other things very well.
- Where
- paragraph 22
- Published
- 2026-10-31
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Unresolved. Evidence up to 2026-11-09.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.