“ChatGPT made all the same kinds of mistakes that its predecessors did.”
Our reading, the sentence that was checked:
ChatGPT made all the same kinds of mistakes that earlier GPT-3 versions did.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Large Language Models like ChatGPT say The Darnedest Things, paragraph 2.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-large-language-models-like-chatgpt-97132a-a1
We recorded 10 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 2023-01-09. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- ChatGPT made
- all the same kinds of mistakes
- that its predecessors did
Our interpretation
- The plain reading concerns persistence of error categories, not identical error rates or identical answers.
- The surrounding examples identify the relevant families: counting and arithmetic, temporal reasoning, physical reasoning, reasoning about people, fabrication, and sexist or racist bias.
- The target is ChatGPT from its launch through the post date. Earlier GPT-3 evidence establishes the comparison baseline and is context rather than evidence of ChatGPT's performance in that period.
- I do not interpret 'all' as requiring an inventory of every conceivable error, but do require coverage of the families identified in the surrounding passage.
Counts as supported if: Independent launch-period records demonstrate persistence of the identified error families, with documented predecessor examples or evaluations establishing that they were already present.
Counts as contradicted if: Comparable tests establish that at least one identified predecessor error family was reliably absent from launch-era ChatGPT. Lower error rates alone do not contradict persistence.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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 horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The evidence supports the persistence of several familiar problems. OpenAI acknowledged erroneous answers at launch, and independent December logs record arithmetic mistakes. Earlier GPT-3 disclosures document untruthful and harmful output. But the retrieved evidence does not establish the full set of error families identified in the passage. This is incomplete support, not evidence that those other failures had disappeared.
Limitations
- Selected failures establish occurrence, not prevalence.
- A weaker claim that some familiar error categories persisted would be supported.
- Improved accuracy would not by itself contradict this claim.
Evidence (3 items)
Supports the claim: OpenAI,
incorrect
Why it matters here: The launch disclosure confirms the persistence of plausible but erroneous output. It supports this part of the claim without establishing every listed error family.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: Vipul Naik, LessWrong,
several mistakes
Why it matters here: The author records sessions on December 21 and 22 using the December 15 ChatGPT version. These logs establish arithmetic and associated reasoning failures, but are selected demonstrations rather than a comprehensive category comparison.
- Publisher
- Vipul Naik, LessWrong
- Kind of source
- Demonstration
- Describes events of
- Public from
Context only: OpenAI,
untruthful, toxic
Why it matters here: This documents relevant failures in preceding GPT-3 models. It supplies part of the historical baseline, not direct evidence about launch-era ChatGPT.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
What would settle it
Independent, dated launch-era records covering the remaining categories, paired with predecessor evaluations or examples under a consistent classification.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Large Language Models like ChatGPT say The Darnedest Things
ChatGPT made all the same kinds of mistakes that its predecessors did.
- Where
- paragraph 2
- Published
- 2023-01-09
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-large-language-models-like-chatgpt-97132a-a1: 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.
About assessment marcus-large-language-models-like-chatgpt-97132a-a1.