Behind the Curtain: How an AI job apocalypse unfolds (www.axios.com)

A two-year-old San Francisco startup called Mercor — founded by three college dropouts and led by a 22-year-old CEO — was just valued at $10 billion after building a business that pays doctors, lawyers and other professionals to train AI so it can perform like human specialists. That deal spotlighted a broader rush to fine-tune large language models and other AI systems with real-world expert labor so machines can take over tasks now done by junior, and eventually senior, employees. High-profile signals from industry leaders — Anthropic’s Dario Amodei warning LLMs could eliminate half of entry‑level white‑collar jobs, Ford’s CEO predicting sweeping white‑collar displacement, and OpenAI hiring former bankers to train models — reinforce that many companies expect greater productivity and profit with much smaller workforces. Technically, this trend centers on creating high-quality, domain-specific training datasets and supervision to push models beyond generic capabilities into expert-level decision-making and analysis. The implication for the AI/ML community is twofold: engineers and product teams will increasingly prioritize fine-tuning, expert labeling, and domain adaptation, while researchers must grapple with robustness, safety, evaluation and the socio-economic impact of automation. Practical takeaways: watch corporate hiring signals, study task-level risk reports (Microsoft’s summer analysis lists interpreters/translators among the most threatened), and prepare for disruptive, uneven labor transitions even as AI drives market gains.