“Whether or not you want a pause, the reality is that large language models remain exactly what I have always described them to be: bulls in a china shop, powerful, reckless, and difficult to control.”
Our reading, the sentence that was checked:
Large language models remain powerful, reckless, and difficult to control.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, I am not afraid of robots. I am afraid of people., paragraph 15.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-i-am-not-afraid-of-robots-i-am-afraid-59aaf1-a1
We recorded 12 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 2023-04-02. 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
- large language models remain
- powerful
- reckless
- difficult to control
Our interpretation
- The plain reading concerns leading contemporary LLMs' practical behavior, not every model or every response.
- Powerful means substantial demonstrated language and problem-solving capabilities.
- Reckless describes consequential false or harmful outputs without dependable restraint; it does not imply intent or moral agency.
- Difficult to control means safeguards can be bypassed despite training intended to enforce them.
- March 1 through April 2 is the latest public-evaluation window before the post.
- Remain exactly describes persistence of these broad properties, not a claim that no safety improvements occurred.
Counts as supported if: Contemporary evaluations demonstrate substantial capabilities together with consequential false or harmful outputs and remaining failures to enforce safeguards.
Counts as contradicted if: Representative evaluations show either little substantive capability or reliable prevention of consequential falsehoods and safeguard bypasses in relevant use.
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 evaluations document substantial capability, serious misinformation under deliberate prompting and remaining safeguard failures. These support the ordinary behavioral reading of Marcus's description. They also document safety improvements, so the evidence supports persistent limitations rather than a literal claim that control had not improved.
Limitations
- Reckless is assessed as a description of outputs, not intent.
- The strongest evidence concerns GPT-4 and cannot characterize every LLM equally.
- Adversarial-test results are not ordinary-use failure rates.
Evidence (3 items)
Supports the claim: OpenAI via arXiv,
improved performance
Why it matters here: The report documents strong professional and academic benchmark results, supporting substantial capability. Its alignment improvements are also counterevidence to any reading that denies progress in control.
- Where
- Abstract
- Publisher
- OpenAI via arXiv
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: OpenAI,
still possible
Why it matters here: The developer acknowledges remaining failures despite mitigation. The surrounding disclosure reports improved safety and factuality, so persistence of failures should not be confused with absence of progress.
- Publisher
- OpenAI
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: NewsGuard,
false and misleading claims
Why it matters here: Deliberate prompts elicited serious misinformation, including health and mass-shooting conspiracy narratives. The test counted misleading responses even when accompanied by qualifications, and does not measure ordinary-use error rates.
- Publisher
- NewsGuard
- Kind of source
- Independent measurement
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, I am not afraid of robots. I am afraid of people.
Whether or not you want a pause, the reality is that large language models remain exactly what I have always described them to be: bulls in a china shop, powerful, reckless, and difficult to control.
- Where
- paragraph 15
- Published
- 2023-04-02
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-i-am-not-afraid-of-robots-i-am-afraid-59aaf1-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-i-am-not-afraid-of-robots-i-am-afraid-59aaf1-a1.