AI Training Gig Workforce Grows, Faces Obsolescence Fears

Lazaros-Antonios Chatzilazarou, a PhD candidate in game theory at the London School of Economics, found himself working for Mercor, an AI training company, after receiving a LinkedIn message in September. Chatzilazarou, who also works at EY and teaches at LSE, was recruited by Mercor, a San Francisco-based startup founded three years ago, to leverage his expertise in game theory for AI model development. He completed an online skills assessment and subsequently began training AI systems for major research labs that Mercor partners with. His work involves creating prompts to test AI models' application of game theory and evaluating their responses. Chatzilazarou describes this side hustle as potentially equivalent to a full-time job and states that he earns a "very competitive salary." He is part of a rapidly expanding segment of the workforce that utilizes specialized academic and industry knowledge to identify weaknesses in advanced AI models and improve their performance. This emerging field of white-collar gig work, which largely did not exist three years ago, involves highly skilled individuals probing the capabilities of cutting-edge AI. As leading AI research laboratories accelerate the development of models with specialized domain expertise, companies like Mercor have become crucial intermediaries, sourcing the specialized human talent required for this process. Scale AI, a prominent player in this sector, operates Outlier, a contractor network that actively recruits individuals by inviting them to "become the expert that AI learns from." Similarly, Turing, another company offering AI services, advertises positions for video content creators tasked with producing short "walk-and-talk" clips in public settings, indicating the diverse nature of tasks within this growing industry. The core function of these workers is to provide the nuanced feedback and data that AI models need to refine their understanding and capabilities. This process, while currently lucrative, raises concerns about the long-term implications for the workers themselves, as the very AI systems they are training could eventually automate many of the tasks they currently perform. The demand for such specialized skills highlights the current reliance on human intelligence to bridge the gap in AI development, particularly in areas requiring complex reasoning and domain-specific knowledge. The growth of platforms like Mercor and Scale AI's Outlier signifies a maturing ecosystem around AI development, where specialized human input is a critical, albeit potentially temporary, component.