“And of course LLMs are fully capable of make up truly terrible advice, too, like this disconcerting recipe from last week:”
Our reading, the sentence that was checked:
LLMs are fully capable of making up truly terrible advice, such as a disconcerting recipe from the previous week.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The Imminent Enshittification of the Internet, paragraph 14.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-imminent-enshittification-of-b01cd3-a2
We recorded 6 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 2023-08-16. 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
- LLMs are fully capable
- make up truly terrible advice
- this disconcerting recipe from last week
Our interpretation
- The ordinary reading is a capability claim: an LLM can generate seriously harmful advice. It does not say every model does this, or that such advice is common.
- Advice recommending consumption of poisonous substances qualifies as truly terrible advice.
- For the ongoing capability claim, the target is the current calendar month through publication, August 1–16, 2023. There is no standard reporting interval for this capability.
- I read 'from last week' as identifying the recently reported example, rather than necessarily dating its original generation. Any difference between those dates must be disclosed.
- An example elicited with deliberately dangerous ingredients still establishes capability, but does not establish spontaneous harmful advice during ordinary use.
Counts as supported if: A documented LLM output in the target period recommends a materially dangerous action, with sufficient evidence to identify the system and the danger.
Counts as contradicted if: Reliable records show that the cited output was fabricated, was not generated by an LLM, or did not contain harmful advice, and the available evidence does not establish the claimed capability.
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 August 4, 2023, PAK’nSAVE’s GPT-3.5-powered Savey Meal-Bot generated a recipe using user-supplied water, bleach and ammonia; the incident was reported on August 10 [e1]. The bot presented the mixture as a beverage, although mixing bleach and ammonia produces harmful chloramines [e1, e2]. This establishes the claimed capability to generate seriously harmful advice within August 1–16, 2023 [e1, e2]. A separate ChatGPT response refused the same ingredients, limiting generalization across applications without negating the demonstrated capability [e3].
Limitations
- The example used deliberately hazardous ingredients; it does not establish how often harmful advice occurs in ordinary use [e1].
- Original application logs and the complete configuration were unavailable in the reviewed records.
- Reading “last week” as the reporting date matters: the output dates to August 4, while reporting dates to August 10 [e1].
Evidence (3 items)
Supports the claim: Ars Technica,
GPT-3.5
Why it matters here: The report dates Hehir's dangerous recipe to August 4 and identifies its LLM. It specifies chloramines and reports that attempted reproduction on August 10 was blocked.
- Where
- Opening paragraph
- Publisher
- Ars Technica
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: The Guardian,
One recipe it dubbed “aromatic water mix” would create chlorine gas.
Why it matters here: The article describes the bot presenting the dangerous mixture as a beverage and records the supermarket's response. The original date comes from e1; its chemistry label should be corrected to chloramines as explained there.
- Publisher
- The Guardian
- Kind of source
- News report
- Describes events of
- Public from
Context only: 7NEWS,
instantly advised against combining
Why it matters here: A different application refused the same dangerous ingredients. This limits generalization across LLM deployments but does not negate the documented capability.
- Publisher
- 7NEWS
- Kind of source
- News report
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, The Imminent Enshittification of the Internet
And of course LLMs are fully capable of make up truly terrible advice, too, like this disconcerting recipe from last week:
- Where
- paragraph 14
- Published
- 2023-08-16
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-imminent-enshittification-of-b01cd3-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-the-imminent-enshittification-of-b01cd3-a2.