I Tracked Down the Hidden Workers Secretly Powering ChatGPT

Worker saw horrific content labeling data for ChatGPT & Gemini — Personal Story (70/100)

At 0:00

👌

Why this score: Credible personal testimony. - Matches extensive reporting on AI data labelers exposed to traumatic content for models like ChatGPT/Gemini. - Video overview confirms this as core thesis with Global South outsourcing and psych toll since 2023.

Original quote: “I definitely worked on ChatGPT. I definitely worked on Gemini. I'm shown, like, honestly, some, like, things that humans should never really witness.”

AI job crisis here — CEOs hide need for hidden human workers — Opinion (50/100)

At 0:11

Setting up the exposé angle — we'll see if it lands 🚩

Why this score: Framing device, not literal claim. - Cites real quotes (e.g., Anderson Cooper interview on AI unemployment hype). - Thesis of 'hidden labor' aligns with verified reports on data labeling workforce; rhetorical contrast is subjective but contextually fair.

Original quote: “[Karen Hao] The AI job crisis is here. But it's not what you've been told. [Anderson Cooper] You've said AI could spike unemployment to 10 to 20% in the next 1 to 5 years. Yes. Tech CEOs talk about a future in which AI systems will soon do away with the need for human labor. But this rhetoric…”

Workers anonymous fearing tech company retaliation — Personal Story (50/100)

At 0:30

Standard NDA fear in this shadowy gig economy 😬

Why this score: Anecdotal report of NDA-enforced anonymity fits known patterns. - Video overview confirms workers often bound by NDAs in AI data labeling. - Common in outsourced Global South roles, per ongoing reports.

Original quote: “Like most people we talked to, she asked to be anonymous for fear that tech companies could retaliate.”

PhD grad applied to 200+ jobs, 3 callbacks; worst market in years — Personal Story (70/100)

At 1:42

200 apps for 3 callbacks? Brutal but tracks with grad hell ⚠️

Why this score: Personal testimony credible amid documented trends. - 2026 grad market toughest in years; 5.6% unemployment > national average. - Over half of grads in non-degree jobs; AI automating entry-level roles.

Original quote: “Over the past year, I've applied to over 200 roles, and I got maybe three callbacks. [Karen] Graduates today are facing the worst job market in years.”

Sources: PhD Degree Holders Are UNEMPLOYED - High Educated and Can't Find Jobs - YouTube

Ivy PhD on food stamps, $15/hr jobs, then $55/hr AI analyst gig — Solid (75/100)

At 2:12

👌 $55/hr for philosophy AI analyst? Real gig tier ✅

Why this score: Anecdote matches market realities. - Specialized AI annotation (e.g., philosophy/RLHF) pays $40-100+/hr; avg $25/hr but PhD roles higher. - Platforms like DataAnnotation/Scale AI list such jobs.

Original quote: “[Karen] To make ends meet, Jen got on food stamps, moved in with her sister, and started working for $15 an hour as a cashier and substitute teacher. Then she saw a job posting on LinkedIn for $55 an hour. [Jen] I think the role I saw was philosophy intelligence analyst.”

Mercor connects OpenAI/Google to workers for AI training — Solid (85/100)

At 2:38

👌

Why this score: Confirmed by multiple sources. - Mercor provides human data/contractors to OpenAI/Google for AI training tasks like data labeling. - Web context verifies it's a real Series C company focused on this.

Original quote: “[3:18] The company is called Mercor, [3:20] and it's one of many contractors which connect companies like OpenAI and Google [3:24] to a distributed workforce to train and improve their AI systems.”

Mercor connects OpenAI/Google to AI training workers — Verified (95/100)

At 2:39

✅👌

Why this score: Confirmed by multiple sources. - Mercor is an AI talent platform founded 2023, explicitly connects contractors to OpenAI/Google for AI training. - $350M Series C in 2025, $1B revenue run rate, 30k+ weekly contractors.

Original quote: “[3:18] The company is called Mercor, [3:20] and it's one of many contractors which connect companies like OpenAI and Google [3:24] to a distributed workforce to train and improve their AI systems.”

AI models progressed from high school to PhD level — Opinion (50/100)

At 3:19

Common industry analogy 🤷‍♂️

Why this score: Subjective benchmark comparison. - 'High school to PhD' is rhetorical framing, not measurable fact. - Reflects genuine capability improvements but oversimplifies complex AI limitations.

Original quote: “[4:20] is that in order to make the next big leap forward [4:23] what they need to do is fit the use case for very specific industries. [4:27] We've gone from maybe the models being like, you know, a [4:30] smart high school student to really starting to get at the PhD level.”

Data workers sourced from Kenya, Venezuela — Solid (85/100)

At 3:30

👌 Known pattern

Why this score: Matches established reporting on AI data labor. - Kenya/Venezuela frequently cited as outsourcing sources for data labeling (Time 2023, NYT reports). - Video overview confirms Global South outsourcing trend for ChatGPT training.

Original quote: “[3:55] For years, tech companies searched for data workers [3:57] largely outside the U.S. [3:59] in low-wage countries like Kenya [4:01] and collapsing economies like Venezuela.”

AI boom drives data center growth and worker demand — Verified (95/100)

At 3:30

✅

Why this score: Directly matches current data. - Over 11,400 data centers globally, projected to triple by 2030 with $7T investment, much AI-driven. - High demand for data workers confirmed in AI supply chain reports.

Original quote: “[3:29] In the same way the AI boom has led to unprecedented growth in data center construction, [3:34] it has also created an insatiable appetite for these workers.”

Data workers sourced from Kenya, Venezuela for low wages — Solid (82/100)

At 3:55

✅

Why this score: Well-documented practice. - OpenAI and others have used workers in Kenya, Venezuela for data tasks due to low costs. - Aligns with video overview and ongoing reports on Global South outsourcing.

Original quote: “[3:55] For years, tech companies searched for data workers [3:57] largely outside the U.S. [3:59] in low-wage countries like Kenya [4:01] and collapsing economies like Venezuela.”

GPT-5 is PhD-level expert — Dubious (45/100)

At 4:09

PhD-level? Pump the brakes 🚩

Why this score: Marketing hype, not technical reality. - GPT-5 (if released by 2026) advances capabilities but no model reaches 'PhD-level expert' across all domains. - Claim echoes industry hype; actual performance varies widely by task/subject.

Original quote: “[4:08] With GPT-5, now it's like talking to an expert, [4:11] a legitimate PhD-level expert that can help you with whatever your goals are.”

Four largest data work startups each make ~$1B/year revenue — Solid (80/100)

At 4:50

👌

Why this score: Claim holds up with industry data. - Data annotation market hit $4.89B in 2025; Scale AI alone projected at $2B in 2026, implying top players near $1B each. - Mercor claims $1B+ run rate, aligning with 'four largest' having ~$1B revenues.

Original quote: “[5:02] By catering to this demand, the four largest data work startups have each had gross revenue of roughly $1 billion a year.”

Scale AI has 700,000+ graduates — Dubious (45/100)

At 5:10

700k? Web says 240k contractors — someone's inflating 💀⚠️

Why this score: Claim appears exaggerated. - Scale AI works with 240,000 gig contractors per recent reports, not 700,000 'graduates'. - 'Graduates' may refer to program completers, but no sources confirm 700k figure; 240k is verified contractor pool. - Possible video-specific claim, but contradicts authoritative data.

Original quote: “[5:10] Scale AI claims to have more than 700,000 graduates at its disposal,”

Jen's Mercor experience: pay drops from $101/hr, contracts end suddenly — Personal Story (50/100)

At 5:21

Rough gig economy reality — race ya for the high-pay tasks 😬💸

Why this score: Personal testimony about Mercor contract instability. - Details pay dropping ($101→$45→$35/hr) and sudden endings match common complaints in AI data labeling work. - *Anecdote can't be independently verified*, but fits broader industry pattern of precarious gig contracts in Global South/AI training.

Original quote: “[5:21] [Jen] We all get a message... My last contract for Mercor offered me $101 an hour. [6:06] I told the teacher... Role ended the next day. [6:19] I shouldn't have to do that. No one should have to do that. Like, just disrupt everything. It's predatory!”

Study: 86% data workers can't meet financial needs, 25% on public aid — Solid (80/100)

At 6:31

👌

Why this score: Stats align with known precarious conditions in AI data work. - Reports describe data labelers facing low pay and insecurity, matching 86% financial struggles and public assistance reliance. - Web context confirms 'digital sweatshops' with exploitation in Global South and US gig economy.

Original quote: “Last year, Tim and his colleagues conducted a study of data workers across the country. They found that 86% struggled to meet their financial responsibilities. A quarter of them relied on public assistance programs like Medicaid and food stamps.”

1 in 5 data workers homeless, median pay under $23k/year — OK (65/100)

At 7:16

Plausible for bottom-end gig, but median feels aggressively low ⚠️

Why this score: Directionally correct but specifics unverifiable. - Web context notes average AI trainer pay ~$65k, but general data labelers (esp. Global South outsourced) often far lower, fitting <$23k median. - Homelessness claim tracks with poverty reports in gig economy; no exact study match but consistent with exploitation narratives.

Original quote: “More than 1 in 5 had experienced homelessness. And overall, the workers reported median earnings of less than $23,000 a year.”

Wang youngest billionaire 2021, unseated last year by Mercor 22yo founders — Verified (95/100)

At 7:56

✅

Why this score: Precisely confirmed by records. - Wang became youngest self-made billionaire in 2021 via Scale AI. - In late 2025 ('last year' from 2026 vid), Mercor founders (22yos) took title with $2B+ net worths post-$10B valuation.

Original quote: “In 2021, Alexandr Wang, the former CEO of Scale, became the world's youngest self-made billionaire. Last year, he was unseated by the three 22-year-old founders of Mercor.”

Ozzy reviewed violent AI-generated gore videos — Personal Story (70/100)

At 8:30

👌

Why this score: Personal testimony aligns with known issues. - Matches reports of data labelers exposed to graphic AI-generated content for training. - Web context confirms human moderators face violent material, leading to trauma.

Original quote: “[Ozzy] It was two dudes murdering a golden retriever with their bare hands. There was, like, weird videos of people constructing, like, furniture out of humans. Like, known celebrities, like, would be, like, in a jail cell, bleeding out, like, with extreme gore.”

Ozzy had nightmares from gore video reviews — Personal Story (70/100)

At 8:51

😔👌

Why this score: Consistent with documented psychological impacts. - AI data workers report PTSD-like symptoms from disturbing content. - Video overview and web context highlight mental health trauma in the industry.

Original quote: “And like, I would have nightmares, honestly, for a couple of weeks after that.”

Unqualified workers do niche tasks like math, counseling — Personal Story (75/100)

At 9:06

👌

Why this score: Anecdote matches industry patterns. - Lawsuits against Surge AI allege workers assigned specialized tasks without qualifications. - Broader reports confirm precarious, mismatched labor in AI data annotation.

Original quote: “He also thought it was strange that he was often asked to perform tasks he was not qualified for. They were just, like, so incredibly niche and specialized. Like, can you help us with this calculus homework, or like, this massive math problem, and then, like, straight to a biology thing. ... I…”

Hundreds of contractors compete for low-cost AI labor — Solid (75/100)

At 10:19

✅

Why this score: Describes real AI supply chain dynamics. - Numerous contract firms vie for work from giants like OpenAI, driving down costs. - Web context confirms competition leads to low wages and poor conditions.

Original quote: “[Tim] There are hundreds of contract companies and they're all working for a small number of top clients at the top of the AI supply chain. It starts at the top, with AI firms seeking to get the best bang for their buck for this kind of labor. To compete for contracts from those AI giants, the data…”

Workers platform-jump, powerless, earn $40 for 13 hours — Personal Story (70/100)

At 10:25

😤

Why this score: Reflects verified exploitation patterns. - Surge AI lawsuit details docked pay, long hours, no benefits. - Industry reports echo low wages (~$2-3/hr) and worker precarity.

Original quote: “When you're platform-jumping all over the place, you feel like you don't have any power or room to stand up and say, “Hey, this isn't right.” I'm working 13 hours a day and I might be making $40.”

Data workers often from vulnerable backgrounds like food stamps, homeless, disabled — Solid (75/100)

At 10:30

👌

Why this score: Plausible and aligns with gig economy patterns. - Gig work targets vulnerable populations due to low barriers and flexible entry. - Video overview confirms poor conditions for Global South data labelers; personal anecdote reinforces. - Web context notes gig growth to $674B in 2026, often for precarious workers.

Original quote: “[10:45] It's not a coincidence that data workers often come from vulnerable backgrounds, [10:49] whether those on food stamps, without housing, or with disability. [10:52] When you're at that point, and I remember being at that point, you take what you can get.”

Data work dismantling full-time jobs, Uber-izing knowledge work per Mary Gray — Opinion (50/100)

At 11:03

Dramatic framing but gig shift is real — not total dismantling tho 🙄

Why this score: Opinion on labor trends with factual basis. - Web context shows gig workers rising (US freelancers 48.5% by 2026), full-time independents doubled 2020-2024. - Mary Gray's 2019 warning on 'Uber-ization' of knowledge work is accurately quoted; aligns with AI gig roles like data labeling. - Exaggerates 'dismantling' — seen as *adding options*, not replacement.

Original quote: “[11:03] It's dismantling full-time employment. [11:06] [Karen] In 2019, a researcher named Mary Gray gave me a haunting warning: [11:11] data work could represent the beginning of the Uber-ization of all knowledge work.”

Silicon Valley sees computers superior, most humans unnecessary — Opinion (50/100)

At 11:43

Acemoglu dropping truth bombs on AI ideology 💥

Why this score: Expert opinion accurately represented. - Daron Acemoglu (2024 Nobel economist) has researched AI's labor impact; web context confirms his focus on automation ideology. - Quote matches his views on elitist attitudes favoring automation over human input. - Contextualized as ideology driving profit, not pure fact.

Original quote: “[11:46] [Daron Acemoglu] The whole ecosystem around Silicon Valley [11:49] is very much based on this idea that computers are superior to humans. [11:54] Most humans are unnecessary.”

AI automation driven by profit + elitist ideology vs workers — Solid (80/100)

At 12:06

✅

Why this score: Credible attribution to expert. - Acemoglu's research critiques AI hype and automation bias; directly supports 'elitist attitude' claim. - Web context affirms his work on labor effects; video overview ties to exploitative practices. - No contradiction; ideological critique is established in econ discourse.

Original quote: “[12:03] He told me that the AI industry's drive for automation [12:06] is fueled not only by profit motives, but also by an ideology. [12:10] There is a sort of elitist attitude towards most workers.”

Employers lay off workers citing AI, then tech firms hire them cheap to train AI — Solid (80/100)

At 12:30

👌

Why this score: Well-documented pattern. - Tech layoffs cite AI efficiency; same firms hire low-wage data labelers for AI training (Global South outsourcing common). - Matches video thesis and web context on exploitation cycle.

Original quote: “[12:53] employers cite AI as a reason to lay off workers. [12:56] Tech firms hire those increasingly desperate workers for cheap — to train AI.”

AI could create unprecedented inequality with corps controlling work — Opinion (50/100)

At 13:05

Dystopian vibes — plausible but speculative 🚩

Why this score: Speculative forecast, not fact. - AI concentrating wealth in tech giants is real concern per reports. - 'Never experienced' and 'completely sidelined' are hyperbolic opinions on future trends.

Original quote: “[13:01] Now, the kind of inequality that we’re talking about here [13:05] could be something we’ve never experienced. [13:08] A handful of corporations controlling most of work, and a large fraction of workers [13:14] essentially completely sidelined from meaningful work.”

AI can augment teachers and nurses instead of replacing them — Solid (85/100)

At 13:38

✅

Why this score: Accurate on augmentation potential. - Pro-teacher AI tools (e.g., MagicSchool) already in use for personalized education. - Pro-nurse AI reduces admin burdens, aids diagnosis despite trust gaps.

Original quote: “[13:23] Rather than automate teaching, [13:25] we can give teachers tools so that they can provide [13:29] individualized education in a much cheaper and effective form. [13:33] We can have pro-nurse AI, meaning [13:36] AI tools that increase the capabilities of nurses for diagnosis, cure, and…”

Turkopticon organizer won changes to Amazon's rejection policy — Personal Story (70/100)

At 14:24

Real organizer story — workers did score wins 👌

Why this score: Credible testimony aligns with known history. - Turkopticon (2010s) was pioneering worker tool for MTurk; fought rejections policy successfully. - Fits video's data worker leverage theme; no contradictions.

Original quote: “[13:57] Krystal was the lead organizer of Turkopticon, one of the first efforts [14:01] to build power to improve labor conditions on Amazon's platform. [14:05] On some issues, they won.”

New California bill AB 2653: Sweatshop-Free AI Procurement Act — Verified (95/100)

At 14:30

✅👌

Why this score: Bill exists and matches description exactly. - AB 2653 amended March 19, 2026, requires certification for AI data work to comply with labor standards, redefines sweatshop labor for AI via Sweatfree AI Code of Conduct. - Introduced 'this year' (2026 video), aligns with timeline.

Original quote: “[14:50] A new bill introduced in California this year [14:53] borrows from that kind of strategy. [14:55] [Assemblymember Lee] This is AB 2653, [14:57] The California Sweatshop-Free AI Procurement Act.”

AB 2653 ensures state AI procurement uses compliant data labor — Solid (85/100)

At 15:00

Spot on 👌

Why this score: Accurate summary of bill's core provision. - Mandates contractors certify data enrichment (labeling, moderation, training) complies with standards like living wage and no psychological harms. - Ties to updated contractor program by July 1, 2027.

Original quote: “[15:00] It would ensure that when the state of California is procuring an AI tool, [15:06] the data work that created those systems actually complied with certain labor standards.”

AI tech not inevitable; we shape its future — Opinion (50/100)

At 15:46

Fair take — policy can steer tech 🚀

Why this score: Opinion on AI development trajectory. - Speaker/Assemblymember argues against inevitability, citing bills like AB 2653 and garment worker models as ways to influence via regulation and coalitions. - Substantiated by real efforts like Newsom's March 2026 AI procurement EO.

Original quote: “[15:21] There is a narrative about how the AI companies operate, [15:24] that this technology is inevitable. [15:27] That is not the case.”