AI training companies are raising billions to get humans to teach chatbots. Here are the startups cashing in. (www.businessinsider.com)

AI’s latest gold rush isn’t new model architectures but the humans who teach them: startups that recruit, vet and pay gig workers to annotate data, rewrite chatbot responses, validate datasets and provide cultural or domain expertise (from memes and Japanese to finance and code) are raising huge rounds and commanding multibillion-dollar valuations. Business Insider highlights players such as Scale AI (over 300,000 gig workers), Surge AI (claims $1.2B revenue, 1M workers and a Forbes-pegged ~$24B valuation), Mercor (raising at ~>$10B, pays ~$95/hr), Micro1 (Series A at $500M valuation, $60M revenue), Turing, Snorkel, Invisible, Labelbox and legacy firm Appen. Reported pay ranges vary widely — from ~$20/hr to specialist rates above $200/hr — and founders and early investors are becoming extremely wealthy even as some firms face layoffs and profitability pressures. Why this matters: human-in-the-loop work (data annotation, RLHF-style feedback, specialized validation and adversarial testing) remains core to improving LLM behavior, safety, and domain competence, creating a booming new labor market and dependence for frontier labs (Anthropic, xAI, Google, Microsoft). The trend amplifies demand for subject-matter trainers as robotics and multimodal models expand, but raises sustainability questions — cost, workforce churn, quality control, and whether these roles will re-automate. The market is therefore shaping incentives for tooling, matching (AI interviews), and platform-level moderation/quality systems that will define how reliably models generalize and stay aligned.