AI Companies Hire Temps to Train Language Models

CBS News reports that leading makers of generative AI are hiring people with diverse skills to train language models, citing job postings and interviews with industry participants. The article profiles Hollywood screenwriter Robin Palmer, who told CBS News she spends 30 hours per week teaching chatbots how to produce creative writing. Christine Cruzvergara, vice president at Handshake, told CBS News these roles are "some of the fastest-growing jobs out there." Brendan Foody, CEO of Mercor, told CBS News "training agents is going to become the largest job category in the world." CBS News also reports that postings often omit the hiring company and require non-disclosure agreements, and it lists sample roles including creative writers, air traffic controllers, litigators, and improv actors.

The article frames these openings as part of post-pretraining work that AI teams use to improve model behavior, which the CBS piece describes as needing "more fine-tuning and reinforcement." Industry-pattern observations: roles focused on dataset curation, structured feedback, and evaluation typically supply the human labels and scenario-specific exemplars used in supervised fine-tuning and human-in-the-loop validation. For practitioners, this means the skills in high demand are both domain expertise and applied annotation/feedback techniques rather than large-scale model engineering alone.

Observers following workforce trends will note two tensions reported by CBS News: high demand for specialized knowledge across many domains, and limited transparency in hiring because postings frequently hide employer identity and require NDAs. Industry-pattern observations: gig and contract models have historically driven rapid, low-friction access to subject-matter expertise but raise questions about pay parity, classification, and attribution for contributed work.

Indicators to monitor include reported hourly rates and contract terms in future postings, any public disclosures about data provenance tied to these engagements, and whether trade groups or regulators propose standards for transparency or worker protections. For practitioners evaluating supplier chains, the prevalence of NDAs and opaque sourcing is a metric for data lineage risk.