“And I think that the fundamental mistake that the field has made, is to assume that capturing one component of intelligence, prediction extraordinarily well, is enough to capture all of intelligence.”
Our reading:
The AI field has mistakenly assumed that capturing prediction well is enough to capture all of intelligence.
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 cited research shows that next-token prediction was already being questioned within AI research. That prevents treating the assumption as uncontested, but does not establish how widely it was held in October 2026. Failures of particular training methods also do not settle whether sufficiently accurate prediction could capture intelligence in principle. Neither component of Marcus's broad attribution is adequately resolved.
Caveats
- A training objective is not a direct measure of researchers' beliefs.
- The admitted paper is historical context and tests a narrower technical proposition.
Evidence (1 item)
Context only: arXiv,
explorations beyond
Why it matters: This research distinguishes inference from training and investigates a planning failure of next-token training. It demonstrates an existing challenge to the paradigm, but neither measures October 2026 consensus nor disproves the sufficiency of an ideal predictor.
- 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- Preprint, not peer reviewed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.
Where we had to interpret
- The plain reading attributes a broadly accepted assumption to the AI field: sufficiently good prediction is sufficient for general intelligence.
- The claim has two empirical components: broad acceptance of that assumption and its inadequacy. Calling it the fundamental mistake adds a judgment about importance.
- Use October 2026 for the alleged ongoing field-wide assumption. Earlier research is historical context.
- Using a prediction objective does not by itself establish a researcher's belief that prediction alone captures every aspect of intelligence.
Right if: Representative research or statements establish broad acceptance of prediction's sufficiency, and suitable evidence establishes a substantive intelligence limitation that remains despite the required predictive competence.
Wrong if: Representative evidence shows that the attributed assumption is not broadly held, or sufficiently comprehensive evidence demonstrates the claimed sufficiency under a clearly stated meaning of intelligence.
We wrote this test after the outcome was known.
What would settle it
Contemporary representative evidence about the field's stated assumptions, together with a precise sufficiency claim and evidence testing that claim.
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
And I think that the fundamental mistake that the field has made, is to assume that capturing one component of intelligence, prediction extraordinarily well, is enough to capture all of intelligence.
- 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.