“The fundamental issue is that chatbots are neither reliable nor truthful, and at the same time don’t really grasp things like human psychology.)”
Our reading, the sentence that was checked:
Chatbots are neither reliable nor truthful and do not really grasp human psychology.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The first known chatbot associated death, paragraph 12.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-first-known-chatbot-associated-88ee66-a1
We recorded 15 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 2023-04-04. 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
- chatbots are neither reliable nor truthful
- don’t really grasp things like human psychology
Our interpretation
- In the surrounding discussion of health advice, read reliable and truthful as practically dependable and factually trustworthy, not as requiring perfect answers or intentional honesty.
- The speaker explicitly acknowledges good advice elsewhere in the passage, so this is not a claim that every chatbot answer is false.
- Operationalize grasping human psychology as robust reasoning about people's beliefs and psychologically sensitive situations. This does not test consciousness or subjective understanding.
- Assess publicly deployed general-purpose LLM chatbots represented by contemporary GPT models, without extending findings to every possible chatbot.
- Use January 1 through April 4, 2023: the latest completed quarter and immediately preceding days for the ongoing state described.
Counts as supported if: Contemporaneous evidence shows material factual or reasoning failures and fragile psychological reasoning that prevent dependable use for the sensitive advice discussed.
Counts as contradicted if: Representative contemporary evaluations establish dependable factual and psychological performance across relevant variations. Isolated successes alone do not contradict a lack of dependability.
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
On the practical reading supplied by the health-advice context, the observation is supported. OpenAI acknowledged factual and reasoning failures in GPT-4 and warned about high-stakes use. Contemporary experiments also found fragile reasoning about people's beliefs in GPT-3.5. Favorable GPT-4 false-belief results matter: they prevent interpreting this assessment as proof that chatbots had no psychological competence or always answered falsely. The supported conclusion is limited to a lack of dependable performance, not a finding about an internal capacity to understand.
Limitations
- The result depends on the operational reading of really grasp.
- The psychological evidence is narrow, model-specific and from preprints.
- No universal claim about every chatbot or every answer is established.
Evidence (3 items)
Supports the claim: OpenAI,
it “hallucinates” facts and makes reasoning errors
Why it matters here: The developer acknowledges factual and reasoning failures and recommends safeguards for high-stakes use. This supports the practical reliability concern, while the same disclosure reports improvements over GPT-3.5.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: arXiv; Tomer D. Ullman,
simple changes to perceptual access confound the model.
Why it matters here: GPT-3.5 failed variants of tasks about what people could see and believe. This supports fragile psychological reasoning, although the author also identifies scene understanding and other reasoning failures as possible explanations.
- Publisher
- arXiv; Tomer D. Ullman
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Context only: arXiv; Michal Kosinski,
solved nearly all the tasks (95%).
Why it matters here: GPT-4 performed well on classic false-belief tasks. This materially challenges a claim of no psychological competence whatsoever, but does not establish robust performance across variations or competence in mental-health advice.
- Publisher
- arXiv; Michal Kosinski
- 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, The first known chatbot associated death
The fundamental issue is that chatbots are neither reliable nor truthful, and at the same time don’t really grasp things like human psychology.)
- Where
- paragraph 12
- Published
- 2023-04-04
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-first-known-chatbot-associated-88ee66-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-the-first-known-chatbot-associated-88ee66-a1.