“As OpenAI has learned, bigger in many ways means better, with outputs more and more humanlike with each iteration. GPT-4 is going to be a monster.”
Our reading, the sentence that was checked:
Because scale improves output in many ways, GPT-4 will be very powerful.
Assessment: Supported Settled before its deadline on 2026-10-29 Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, What to Expect When You’re Expecting … GPT-4, paragraph 32.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-what-to-expect-when-youre-expecting-651690-a1
We recorded 21 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 2022-12-25. 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
- Settled before its deadline
- Deadline state on 2026-10-29
- Settled before its deadline (the release’s reference date)
From the speaker
- bigger in many ways means better
- outputs more and more humanlike with each iteration
- GPT-4 is going to be a monster
Our interpretation
- The ordinary capability forecast is that GPT-4 will be a substantial, broadly capable advance over GPT-3 and the GPT-3.5 system underlying contemporary ChatGPT.
- Human judgments of response quality and varied task evaluations operationalize the claimed improvement. They do not establish human understanding or indistinguishability.
- Scaling is the stated rationale. This assessment tests the capability forecast and treats scaling studies as context; it does not claim to identify scaling's isolated causal contribution.
- The surrounding warnings rule out interpreting monster as flawless performance.
Counts as supported if: Release-model evaluations show substantial improvements over predecessors across varied tasks and in human judgments of useful language outputs.
Counts as contradicted if: Broad comparative evaluations show no substantial capability improvement, or widespread regression, relative to GPT-3 and GPT-3.5.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Did the forecast come true?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
- Settled before its deadline
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Under the ordinary capability reading, the forecast is supported. The original GPT-4 report records substantial task improvements and a clear human preference for its responses over GPT-3.5. Early-model experiments corroborate the breadth of the advance. Scaling research supports the general rationale, but these records do not establish that size alone caused GPT-4's gains or that its outputs became human in every relevant sense.
Limitations
- The decisive evaluation was conducted by the model developer.
- Microsoft's corroborating study used an early model and came from an OpenAI partner.
- The outcome depends on reading monster as a broad capability advance, not as a precise causal or human-indistinguishability claim.
- Capability improvements need not occur on every task.
Evidence (4 items)
Supports the claim: OpenAI / arXiv,
the ability to follow user intent
Why it matters here: The report records preference for GPT-4 on 70.2% of 5,214 prompts and broad benchmark gains over predecessors. These are developer-run evaluations supporting a substantial capability and response-quality advance, not proof of human understanding.
- Publisher
- OpenAI / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: Microsoft Research / arXiv,
GPT-4's performance
Why it matters here: The authors report experiments across mathematics, coding, vision, medicine, law, and psychology. Because they examined an early development version, this corroborates the capability direction but is not treated as a matched evaluation of the released model.
- Where
- Abstract
- Publisher
- Microsoft Research / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: OpenAI researchers / arXiv,
a fixed compute budget.
Why it matters here: The study reports predictable relationships between language-model loss, model size, data, and compute. It supports the general scaling rationale but does not isolate why GPT-4 improved.
- Where
- Abstract
- Publisher
- OpenAI researchers / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: DeepMind researchers / arXiv,
substantially less compute
Why it matters here: The study found that a 70-billion-parameter model trained on more data outperformed several larger models. This qualifies a parameter-count-only reading of bigger means better; it does not contradict improvement from jointly scaling resources.
- Where
- Abstract
- Publisher
- DeepMind researchers / arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, What to Expect When You’re Expecting … GPT-4
As OpenAI has learned, bigger in many ways means better, with outputs more and more humanlike with each iteration. GPT-4 is going to be a monster.
- Where
- paragraph 32
- Published
- 2022-12-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-to-expect-when-youre-expecting-651690-a1: Current. Outcome: Supported. 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.
About assessment marcus-what-to-expect-when-youre-expecting-651690-a1.