Hollywood Writers Shift to AI Training: First-Person Account of Data Annotation Work

A first-person account from a Hollywood writer and showrunner — author of shows on Paramount, Hulu, and BBC — describes their transition into AI training work after the 2023 writers' strike dried up traditional income. They now work under a handle like 'ri611' performing data annotation tasks for companies including Mercor, Outlier, Task-ify, Turing, Handshake, and Micro1.

How They Got In

• Discovered the opportunity via an unofficial Writers Guild Facebook group: a post mentioned Mercor paying $150/hour for writers.
• Applied to 10 jobs, spent 20 unpaid hours on tests, interviewed with an AI recruiter agent (a flickering light on screen).
• Hired six weeks later as a 'generalist' data annotator at $52/hour — below 'expert' but above entry level.

Daily Tasks

• Read conversations between users and a major LLM chatbot, grading responses on a scale of 1–5, with written justification.
• Assess tone: natural vs flat, affected vs annoying.
• Annotate images (furniture patterns, group photos to isolate individuals).
• Time-stamp video events: dog barking, a stranger walking past a window, a balloon popping.
• Generate sensitive content (anime sex scenes, violent imagery, bomb recipes) as part of red-team safety testing.

Working Conditions

• Tasked via Slack channels, Airtables, payment portals, and Google Workspace apps.
• A team leader explicitly stated: 'These are not jobs, these are tasks, and we are taskers.'
• Project manager: a 22-year-old recent grad who intended to go into investment banking.
• Pays rent, buys food, covers a human cleaner ($150 flat rate).

Context for Developers

This account illustrates the human labor behind LLM training data. Writers — who are experts in tone, narrative, and safety — are now grading chatbot responses and generating edge-case content. For AI engineers, it's a reminder that data quality depends on underpaid, precarious contractors. Tools like Outlier and Mercor broker this labor, and the task structure (per-task, no benefits) mirrors gig platforms.