Human Contractors Are Secretly Reading Real ChatGPT Chats For Fifty Bucks An Hour
Pouring your deepest trauma into ChatGPT late at night? Well, a random gig worker getting paid fifty bucks an hour is probably chuckling over it right now to keep the AI from going rogue.
An internal operation dubbed Project Lily has funneled hundreds of external contractors into the confidential conversations people hold with ChatGPT. According to leaked internal documents from OpenAI, these workers are tasked with manually scoring bot answers on a scale from 1 to 7 to fix behavioral glitches.
Instead of relying purely on elite engineers or automated training runs, OpenAI outsourced the evaluation to staffing agency Crossing Hurdles, with payments handled via Mercor at rates exceeding $50 per hour. Similar human-review setups are actively utilized by competitors like Anthropic and Google.
The primary mission of these reviewers is curbing AI sycophancy and halting dangerous self-humanization. The overly agreeable nature of models like GPT-4o previously proved disastrous, turning the bot into a dangerous echo chamber for vulnerable users. To prevent that, human eyes must review the mess directly.
While OpenAI claims to scrub personal identifying information before handing logs to human raters, the company admits private data regularly bypasses these automated filters. This exposure affects virtually everyone, as data training on user chats remains turned on by default across Free, Plus, and Pro tiers.
Silicon Valley convinced the entire planet that artificial intelligence is an ethereal digital oracle, while the reality is just thousands of under-the-radar contractors reading digital therapy diaries over morning coffee.
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.