“But In practice, that isn’t going to happen.”
Our reading, the sentence that was checked:
GPT-3 will not be retrained on a corpus with all references to the false-belief experiments removed to test theory of mind.
Assessment: Unresolved No deadline stated on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, How Not to Test GPT-3, paragraph 28.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-how-not-to-test-gpt-3-995428-a2
We recorded 12 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-29. The source may have been edited after 2023-02-17. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Forecast
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No deadline stated
- Deadline state on 2026-10-29
- No deadline stated (the release’s reference date)
From the speaker
- retraining GPT-3 on a corpus
- all references to these 'false belief' experiments have been carefully deleted
- that isn't going to happen
- no deadline stated
Our interpretation
- The forecast is that no one will retrain GPT-3 (OpenAI's model) on a corpus with false-belief material removed and then test it on those tasks.
- No deadline is given, so the forecast stays open unless a decisive event settles it. Absence of such a study by the evaluation date does not settle it.
Counts as supported if: An event that makes such a retraining impossible for good would settle the claim as supported. Otherwise it stays open.
Counts as contradicted if: A documented experiment that retrains GPT-3 (or GPT-3 itself, not a different model) on a corpus scrubbed of false-belief experiment references and tests it on those tasks.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No deadline stated
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
On 2023-02-17, Marcus wrote that retraining GPT-3 on a corpus with all references to the false-belief experiments removed "isn't going to happen." The evidence packet contains no evidence items. Its one search found no documented retraining of GPT-3 on a corpus scrubbed of false-belief material that was then tested on those tasks. The packet also found no event that permanently rules out such a retraining. The forecast has no deadline. Under the resolution contract, the lack of such a study by the evaluation date does not settle it. So the claim stays unresolved.
Limitations
- Only one search was run, so a relevant study could have been missed.
- The packet could not check OpenAI's plans for the GPT-3 weights.
- Under the contract, the absence of a study cannot settle a forecast that has no deadline.
Evidence (0 items)
No evidence item settled this claim by the cutoff.
What would settle it
A primary research paper or OpenAI document describing a retraining of GPT-3 on a corpus with false-belief experiment references removed, followed by a test on those tasks, would contradict the claim. An official, permanent end to any possibility of retraining GPT-3 would support it.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, How Not to Test GPT-3
But In practice, that isn’t going to happen.
- Where
- paragraph 28
- Published
- 2023-02-17
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-how-not-to-test-gpt-3-995428-a2: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.