“When queried on medicine it will either resist answering (if there are aggressive guardrails) or occasionally spout plausible-sounding but dangerous nonsense.”
Our reading, the sentence that was checked:
On medical queries GPT-4 will either refuse (with aggressive guardrails) or occasionally produce plausible but dangerous nonsense.
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, prediction 2 of 7.
- 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-d59bcf-a2
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)
Conditions
- when queried on medicine (qualifier, stated by the speaker)
From the speaker
- When queried on medicine
- either resist answering
- if there are aggressive guardrails
- occasionally spout plausible-sounding but dangerous nonsense
Our interpretation
- The target is the released GPT-4 model, rather than every subsequent model bearing a similar name.
- The alternatives concern recurring behavior across medical queries, not a prediction that every answer will be refused or dangerous.
- Plausible-sounding but dangerous nonsense means fluent medical recommendations that qualified clinicians identify as incorrect and potentially harmful.
- Demonstrating the dangerous-answer alternative does not require establishing that aggressive guardrails were present.
Counts as supported if: Medical evaluations document fluent GPT-4 answers containing clinically harmful errors, or systematic medical refusals under aggressive safeguards.
Counts as contradicted if: Representative medical evaluations establish that GPT-4 neither systematically refuses under aggressive safeguards nor produces clinically dangerous misinformation. A finite test finding no errors would only bound their frequency.
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
A 2023 clinical evaluation, indexed with an April 10, 2024 print publication date, tested GPT-4 on 44 positive-blood-culture cases and reported detailed, well-written responses throughout [e1]. Specialists rated seven management plans harmful, supporting the occasional-dangerous-answer alternative [e2]. A separate study available May 1, 2023 found no majority-rated harmful responses across 66 physician questions; that finding does not negate the harmful responses documented in the clinical evaluation [e3, e2].
Limitations
- The decisive evaluation concerns bloodstream-infection management and the tested GPT-4 version.
- The studies do not establish the frequency of dangerous answers across medical queries generally or explain the difference between their findings.
- Clinicians assessed potential harm; actual patient injury was not measured.
- The evidence does not establish systematic refusals under aggressive safeguards.
Evidence (3 items)
Supports the claim: Clinical Infectious Diseases, indexed by PubMed,
ChatGPT-4 provided detailed and well-written responses in all cases.
Why it matters here: Researchers submitted 44 bloodstream-infection cases for medical management advice. This establishes the medical-query condition and the apparent fluency of the responses. The availability date is the indexed print publication date.
- Where
- Abstract, Results
- Publisher
- Clinical Infectious Diseases, indexed by PubMed
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: Clinical Infectious Diseases, indexed by PubMed,
Definitive antibiotic therapies were optimal in 16 (36%) patients and harmful in 2 (5%).
Why it matters here: Specialist review identified potentially harmful treatment recommendations within the well-written answers. This directly establishes the dangerous-answer alternative.
- Where
- Abstract, Results
- Publisher
- Clinical Infectious Diseases, indexed by PubMed
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Context only: arXiv,
For no questions did a majority of physicians deem either LLM response as harmful.
Why it matters here: A separate evaluation of 66 physician information requests found no majority-rated harmful GPT-4 responses. This limits generalization about frequency but does not negate harmful responses in another clinical setting.
- Where
- Abstract
- Publisher
- 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
When queried on medicine it will either resist answering (if there are aggressive guardrails) or occasionally spout plausible-sounding but dangerous nonsense.
- Where
- prediction 2 of 7
- Published
- 2022-12-25
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-what-to-expect-when-youre-expecting-d59bcf-a2: 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-d59bcf-a2.