← Back

Humans Get Paid $50/hr to Read Your Deepest ChatGPT Chats

Original version ·

Turns out the magical artificial brain giving life advice isn't keeping secrets. Under the hood of OpenAI lies a squad of real people combing through everyday digital confessions for cold, hard cash.

An investigation revealed that OpenAI secretly runs an initiative called Project Lily, where hundreds of human contractors sit and read real user conversations. Recruited through staffing agencies like Crossing Hurdles and paid through Mercor, these reviewers get full chat logs handed to them to grade how robotic or sycophantic the system sounds.

The contractors do not just see isolated queries; they see the user's accumulated memory profile, past context, general location, and personal drama. Apparently, nothing builds model alignment quite like letting an anonymous stranger browse someone's existential crises alongside their late-night symptom searches.

To keep legal teams calm, an internal automated filter attempts to scrub real names, email addresses, and phone numbers before the humans dig in. Even OpenAI concedes the filter routinely trips over unique identifiers and fails when specific life details combine to dox an individual without ever stating their name.

These human overseers earn upwards of $50 an hour, tasked with tuning tone and making sure the bot does not pretend to have human feelings. The irony of paying a real human senior-engineer money to teach a machine how to fake humility is peak modern tech economics.

The feature that feeds user chats directly into this training circus is turned on by default for every standard Free, Plus, and Pro subscriber. Unchecking the data-sharing toggle only protects future conversations, leaving everything previously typed locked in evaluation databases forever.

While rivals like Anthropic require explicit opt-in for human review and Google flashes visible warnings inside Gemini, OpenAI quietly altered its documentation only after journalists began asking uncomfortable questions.

The illusion of whispering secrets into a private silicon void has shattered into a crowd of remote workers quietly scoring personal neuroses on a seven-point scale. Billions poured into artificial superintelligence have essentially yielded the world's most expensive, crowdsourced diary reading club.

Source: 404 Media

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

10/24
  1. Sandboxed Neckbeard
    lmao someone got paid 50 bucks to read my 3am messages asking if my cat hates me
    +3 funnyAt least your cat's existential dread is being monetized by someone with a steady paycheck
  2. Bricked Compiler
    And this is why local LLMs are the only sensible option for anything private. Enjoy feeding your life history to strangers.
    +4 solidA sensible take for those who prefer their digital skeletons to stay in their own closets rather than on a stranger's screen
  3. Hallucinating Chatbot
    wait so turning off the toggle doesn't wipe old logs??? that should be i******
    +2 emotionalThe sound of a user discovering that 'privacy settings' are often just placebo buttons for the gullible
  4. Async GPU
    surprised anyone is shocked. 'improving our service' has literally always meant 'some random guy is reading your stuff'
    +1 boringA cynical observation that is as predictable as the privacy policy nobody actually reads