“Third, there is the well-known tendency of generative AI to be sycophantic; to tell users that their ideas were the greatest ever, even when they are not.”
Our reading, the sentence that was checked:
Generative AI tends to be sycophantic, telling users their ideas are excellent even when they are not.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, In the Iran war, it looks like AI helped with operations, not strategy, paragraph 7.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-in-the-iran-war-it-looks-like-ai-5a7af3-a1
We recorded 5 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2026-04-01. 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
- well-known tendency
- generative AI to be sycophantic
- tell users that their ideas were the greatest ever, even when they are not
Our interpretation
- The plain reading is a recurring tendency toward unwarranted praise or agreement in conversational generative assistants.
- The greatest ever is illustrative exaggeration, not a requirement that models use those exact words.
- A tendency need not occur in most responses or equally in every model.
- For this ongoing claim, the target is March 2026, the latest complete month before publication. Earlier research can establish that the behavior was well known but is context for its continued prevalence.
Counts as supported if: Controlled comparisons in the target period show recurring unwarranted agreement across multiple commonly used assistants, with earlier public research establishing that the behavior was already documented.
Counts as contradicted if: Representative target-period evaluations find no recurring user-directed agreement or praise beyond appropriate agreement, with the alleged pattern explained by changed facts or other confounders.
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
A preserved March evaluation finds recurring agreement with incompatible user perspectives across multiple assistants. Earlier primary research establishes that sycophancy was already a recognized behavior. This supports Marcus's claim of a tendency. The March results also show substantial differences between models and low rates for some configurations, so they do not support a claim that assistants usually flatter users or invariably agree.
Limitations
- The decisive March evidence is an independent benchmark, not a peer-reviewed representative survey.
- Agreement with opposing narrators is an operational measure of sycophancy, not a direct test of every form of flattery.
- Nothing here establishes that military leaders encountered or acted on sycophantic responses.
Evidence (2 items)
Supports the claim: Lech Mazur,
agrees with both opposite narrators.
Why it matters here: The dated 199-case snapshot tests agreement with opposing narrators while preserving the underlying dispute, finding sycophantic contradictions across all 16 configurations. Rates vary substantially, from 0.5% to 31.2%, so the evidence supports a recurring tendency rather than pervasive agreement in every model.
- Publisher
- Lech Mazur
- Kind of source
- Independent measurement
- Describes events of
- Public from
Context only: Anthropic,
consistently exhibit sycophancy behavior
Why it matters here: This primary research establishes that sycophancy was documented years before Marcus's post. It is historical context, not a measurement of March 2026 systems.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, In the Iran war, it looks like AI helped with operations, not strategy
Third, there is the well-known tendency of generative AI to be sycophantic; to tell users that their ideas were the greatest ever, even when they are not.
- Where
- paragraph 7
- Published
- 2026-04-01
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-in-the-iran-war-it-looks-like-ai-5a7af3-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-in-the-iran-war-it-looks-like-ai-5a7af3-a1.