“A bunch of that work was done poorly paid labor in Kenya, paid less than $2/hour to evaluate (e.g.) graphic descriptions of sexual situations involving children and animals that I prefer not to describe it in detail.”
Our reading, the sentence that was checked:
A bunch of the human filtering work behind ChatGPT was done by poorly paid workers in Kenya, paid less than $2/hour to evaluate graphic material such as descriptions of sexual situations involving children and animals.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, 24 Seriously Embarrassing Hours for AI, paragraph 9.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-24-seriously-embarrassing-hours-for-cd2743-a3
We recorded 8 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 2023-01-18. 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)
Conditions
- a bunch of that work (scope, stated by the speaker)
From the speaker
- a bunch of that work
- poorly paid labor in Kenya
- paid less than $2/hour
- to evaluate (e.g.) graphic descriptions of sexual situations involving children and animals
Our interpretation
- Plain reading: a meaningful part of the human filtering work for ChatGPT was done by Kenyan workers who took home under about $2 per hour and labelled graphic text, including child sexual abuse and bestiality.
- Period: the Sama text-labelling contract for OpenAI, from November 2021 to March 2022.
- 'Less than $2/hour' is read as take-home pay up to about $2, matching the TIME headline he cites.
Counts as supported if: Reliable reporting shows Kenyan contractors labelled graphic sexual content involving children and animals for OpenAI's ChatGPT safety work, for take-home pay of about $2/hour or less.
Counts as contradicted if: Reliable evidence shows the workers were paid well above $2/hour, did not label such material, or that the work was unrelated to ChatGPT.
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
TIME published its report on 2023-01-18. It says that, starting in November 2021, OpenAI sent tens of thousands of text snippets to an outsourcing firm in Kenya to get them labelled. Some of those snippets described child sexual abuse and bestiality in graphic detail (e2). TIME reviewed internal Sama and OpenAI documents, including payslips, and interviewed four Sama employees who worked on the project. It reports that the labellers Sama employed for OpenAI took home about $1.32 to $2 per hour, depending on seniority and performance (e1). So the top of that range is $2, and the rest is below it. A Sama spokesperson said workers could earn $1.46 to $3.74 per hour after taxes (e3). The pay and tasks TIME reports match the claim that a bunch of the work was done by Kenyan workers paid under $2 per hour to label this kind of material.
Limitations
- All the evidence comes from one investigative report, TIME's article of 2023-01-18. No independent source in the packet confirms the pay figures or what the tasks involved.
- The quoted passages describe labelling work done for OpenAI. They do not by themselves spell out how that work fed into ChatGPT.
- TIME's range goes up to about $2 per hour, which is not strictly less than $2.
- Sama gave a higher range of $1.46 to $3.74 per hour, and the packet notes it was not broken down by role. It cannot be fully squared with TIME's payslip-based figures.
- The evidence does not show what share of all of ChatGPT's human filtering work these Kenyan workers did.
Evidence (3 items)
Supports the claim: TIME,
The data labelers employed by Sama on behalf of OpenAI were paid a take-home wage of between around $1.32 and $2 per hour depending on seniority and performance.
Why it matters here: Based on payslips, Kenyan labellers took home about $1.32–$2 per hour, which matches 'less than $2/hour'.
- Where
- paragraph 8
- Publisher
- TIME
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: TIME,
Some of it described situations in graphic detail like child sexual abuse, bestiality, murder, suicide, torture, self harm, and incest.
Why it matters here: Confirms the Kenyan workers labelled graphic descriptions of child sexual abuse and bestiality.
- Where
- paragraph 6
- Publisher
- TIME
- Kind of source
- News report
- Describes events of
- Public from
Context only: TIME,
workers could earn between $1.46 and $3.74 per hour after taxes.
Why it matters here: Sama disputed the pay range, saying some workers could earn above $2. Sama would not say which roles earned at the top of the range.
- Publisher
- TIME
- 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, 24 Seriously Embarrassing Hours for AI
A bunch of that work was done poorly paid labor in Kenya, paid less than $2/hour to evaluate (e.g.) graphic descriptions of sexual situations involving children and animals that I prefer not to describe it in detail.
- Where
- paragraph 9
- Published
- 2023-01-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-24-seriously-embarrassing-hours-for-cd2743-a3: 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-24-seriously-embarrassing-hours-for-cd2743-a3.
HeldBy this site
The final-text check failed: rationale_supported_by_evidence; overclaims: OpenAI's statement to TIME confirmed only that Sama workers in Kenya contributed to a toxic-content detector later built into ChatGPT..
About assessment marcus-24-seriously-embarrassing-hours-for-cd2743-a2, which was not published.