AI Training CEO Declares End of Data-Labeling Era, Emphasizes Need for Real-World Expertise and Research-Driven Partnerships
"The era of data-labeling companies is over," declared Jonathan Siddharth, CEO of Turing, a $2.2 billion AI training firm, in a recent appearance on the "20VC" podcast. Siddharth argued that the traditional model of simple data annotation—such as tagging images or categorizing text—is no longer sufficient for the demands of modern artificial intelligence. He explained that while early AI models relied heavily on basic labeling tasks that could be outsourced at scale, today’s advanced systems—particularly agentic models and those built with reinforcement learning—require far more sophisticated data. "Data needs have significantly changed," Siddharth said. "It's no longer just about labeling; it's about real-world data that reflects how humans actually perform knowledge work across industries." According to Siddharth, the future lies in creating simulated environments—mini-worlds that replicate complex human workflows in fields like healthcare, finance, and engineering. These environments allow AI systems to learn through interaction and decision-making, much like humans do. To build them, AI training firms must now recruit domain experts, not just general annotators. "This is now the era of research accelerators," Siddharth said, emphasizing that top AI labs are seeking partners who can act as proactive collaborators in research, not just task processors. Turing recently raised $111 million in Series E funding, bringing its valuation to $2.2 billion. The company reported a 2024 annual revenue run rate of $300 million—nearly triple its previous year’s figure. The rise of AI training has fueled a booming market for data labeling, attracting massive investments. In June, Meta acquired a 49% stake in Scale AI, valuing the company at over $29 billion. Earlier in the year, Mercor secured funding that valued it at $10 billion. This surge has also led to a rapid expansion of a freelance workforce, with some contractors earning thousands of dollars monthly. However, the work can be emotionally taxing and inconsistent, as reported by Business Insider, which interviewed over 60 data labelers. Compounding the issue is an underground market for access to training platforms. Business Insider recently uncovered more than 100 Facebook groups selling unauthorized access to both real and fake contractor accounts. Despite strict policies against such practices, scammers and opportunists are exploiting the high demand for AI training gigs.
