“I think it’s fair to say that they have not. GPT-Turbo released this week still produces its share of groaners, like this interchange Phil Libin just sent:”
Our reading, the sentence that was checked:
Hallucinations and stupid errors in LLMs have not been solved; GPT-4 Turbo still makes them.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Evidence that LLMs are reaching a point of diminishing returns - and what that might mean, paragraph 21.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-evidence-that-llms-are-reaching-a-7cbffe-a1
We recorded 20 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 2024-04-13. 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
- they have not
- hallucinations and stupid errors
- GPT-Turbo released this week
- still produces its share of groaners
Our interpretation
- The ordinary reading is that factual mistakes and plainly unsuccessful responses persisted, including in the newly released GPT-4 Turbo. It does not claim that accuracy had never improved.
- Interpret GPT-Turbo as gpt-4-turbo-2024-04-09 and assess its release through the April 13 post.
- The claim supplies no error-rate threshold. Documented failures can establish persistence without establishing their prevalence.
Counts as supported if: Dated tests or reproducible reports identify continuing factual mistakes and task failures in the April GPT-4 Turbo release.
Counts as contradicted if: Credible evidence establishes that the alleged failures were misattributed or erroneous and that the named release had eliminated the relevant problems across its intended uses. Isolated successful answers are insufficient.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
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
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
Contemporary independent coding tests found failures in the April 9 Turbo release. A dated primary user report also identified incorrect factual self-description and supplied reproduction code. Together they support the limited observation that errors persisted. They do not establish how often ordinary users encountered hallucinations or validate every detail of the unseen Libin example.
Limitations
- The coding test measures task completion, not hallucination frequency.
- The factual-error evidence is a user report with reproduction instructions, not independent replication.
- Persistence does not establish worsening performance or an inability to improve.
Evidence (2 items)
Supports the claim: Aider,
62%
Why it matters here: The tests used the newly released snapshot on 133 Python exercises, allowing two attempts. Unsuccessful tasks demonstrate that coding errors persisted; this is not a measurement of factual hallucinations.
- Publisher
- Aider
- Kind of source
- Benchmark result
- Describes events of
- Public from
Supports the claim: FlyingFathead, OpenAI Developer Community,
2021.
Why it matters here: This primary user report documents incorrect factual self-description in the named snapshot and supplies reproduction code. It is a reported demonstration, not an OpenAI acknowledgment or an independently replicated error-rate study.
- Publisher
- FlyingFathead, OpenAI Developer Community
- Kind of source
- Demonstration
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Evidence that LLMs are reaching a point of diminishing returns - and what that might mean
I think it’s fair to say that they have not. GPT-Turbo released this week still produces its share of groaners, like this interchange Phil Libin just sent:
- Where
- paragraph 21
- Published
- 2024-04-13
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-evidence-that-llms-are-reaching-a-7cbffe-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-evidence-that-llms-are-reaching-a-7cbffe-a1.