“It’s just a matter before people start to realize that we need some genuinely new ideas in the field, either new mechanisms (perhaps neurosymbolic2), or different approaches altogether.”
Our reading:
People will eventually realize that AI needs genuinely new ideas, such as neurosymbolic mechanisms or different approaches.
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
On 2025-11-25, Sutskever predicted a return to experimental research and described models' weak generalization as a fundamental problem [e1]. His 2023-03-27 interview already discussed techniques that could improve reasoning alongside scaling [e2]. Meta had presented LeCun's proposed alternative architecture on 2022-02-23 [e3]. These records do not establish that AI participants began to recognize, after 2023-09-22, a need for genuinely new ideas to overcome the discussed limitations [e1, e2, e3]. They also do not rule out that recognition emerging later [e1, e2, e3].
Caveats
- The evidence does not establish a change from earlier views to recognition that new mechanisms or different approaches are necessary.
- Later discussion of research does not by itself show that this recognition emerged or spread.
- The records do not establish whether new approaches are technically necessary.
- The forecast has no deadline or numerical threshold for recognition.
Evidence (3 items)
Context only: Dwarkesh Podcast,
a return
Why it matters: In this original interview, Sutskever describes a return to experimental research after emphasis on scaling. This is suggestive of the forecast, but part of his description is itself prospective and does not establish an independently corroborated change in people's recognition of necessity.
- Publisher
- Dwarkesh Podcast
- Kind of source
- News report
- About events on
- Published
Context only: Dwarkesh Podcast,
things that help
Why it matters: Before Marcus's post, Sutskever already discussed additional techniques, including reasoning tokens, alongside scaling. The earlier and later interviews suggest a change of emphasis but do not cleanly establish when he newly accepted the necessity specified by the forecast.
- Publisher
- Dwarkesh Podcast
- Kind of source
- News report
- About events on
- Published
Context only: Meta AI,
six separate modules
Why it matters: Meta presented LeCun's alternative architecture before the forecast. Later repetition of that research agenda would not by itself demonstrate new recognition after September 2023.
- Publisher
- Meta AI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- 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 ordinary reading predicts emerging recognition after publication that overcoming the discussed limitations requires new mechanisms or different approaches.
- In context, 'people' means participants in the AI field. 'Start to realize' requires evidence of recognition emerging or spreading after 2023-09-22, rather than merely finding later statements expressing an existing view.
- The claim does not require majority agreement, abandonment of neural networks, or adoption of neurosymbolic AI specifically.
- Assess recognition of a need, rather than whether that need is technically proven. Research on new techniques alone does not establish such recognition.
- No numerical threshold is supplied. A credible account of changing views, supported by independent participants' records, or comparable observations over time could establish the predicted development.
Right if: Evidence documents recognition emerging or spreading after 2023-09-22 among AI participants that genuinely new mechanisms or different approaches are needed. It must establish change over time and acceptance of necessity, rather than merely interest in innovation.
Wrong if: A decisive event would have to rule out the eventual emergence of this recognition. Continuing disagreement, technical progress, or failure to observe the change by the evaluation date does not contradict an unlimited forecast.
We wrote this test after the outcome was known.
What would settle it
Comparable dated statements or independently corroborated accounts showing that recognition emerged or spread after 2023-09-22 among AI participants that genuinely new ideas were needed to overcome the discussed limitations. The evidence must show changing recognition of necessity. Later agreement or research activity alone is insufficient; majority agreement and endorsement of neurosymbolic AI are not required.
We found 10 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, Elegant and powerful new result that seriously undermines large language models, paragraph 25.
- 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 2023-09-22.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, Elegant and powerful new result that seriously undermines large language models
It’s just a matter before people start to realize that we need some genuinely new ideas in the field, either new mechanisms (perhaps neurosymbolic2), or different approaches altogether.
- Where
- paragraph 25
- Published
- 2023-09-22
- 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: the write-up said more than the evidence showed, and our summary didn’t match their words.
Held backBy us
A reviewer said our test wasn’t fair to what was actually said.