AI Giants Shift to Expert-Labeled Data for Advanced AI Models

- AI leaders like Scale AI, Turing AI, and Toloka are replacing low-cost labelers with domain experts. - Experts in fields such as finance and biology now earn 20-30% more to create advanced datasets. On July 20, 2025, Mitrade reported that leading AI companies, including Scale AI and Turing AI, are making a significant change by transitioning from low-cost data labelers in regions like Africa and Asia to highly paid specialists from fields such as finance, biology, and engineering. This pivotal shift is driven by the industry's increasing demand for expert-driven datasets to train sophisticated AI systems like OpenAI's o3 and Google's Gemini 2.5, as these advanced models prioritize reasoning and nuanced insights that simple data annotations cannot provide. Consequently, the evolution of “reasoning” AI has caused companies like OpenAI, Google, and Meta to reframe their priorities, and they now aim to refine general AI functionality and reliability. These advanced systems depend on intricate, domain-specific training data that traditional, low-cost annotations often fail to provide. Therefore, AI executives believe that a qualitative leap in dataset construction is necessary to resolve challenges involving human reasoning and bias. Several industry leaders have stepped up to facilitate this shift. In June, Meta boosted Scale AI’s funding, raising its valuation to $29 billion and purchasing a minority stake in the company. Turing AI, widely recognized for its contributions to OpenAI’s code, secured $111 million in a Series E funding round in March, bringing its valuation to $2.2 billion. Similarly, Toloka raised $72 million through Bezos Expeditions in May, focusing on enhancing hybrid human-AI systems to meet these emerging demands. This deeper commitment to data quality is also reshaping compensation. According to the report, Jonathan Siddharth, CEO of Turing AI, disclosed that his company pays domain experts 20-30% more than current market rates, reflecting the premium on specialized expertise. In addition, companies are allocating more of their budgets to acquiring high-quality data, with many now dedicating 10-15% of their resources to create these training datasets. These developments confirm the industry now widely acknowledges data's central role in AI training. The Mitrade report cited Olga Megorskaya, CEO of Toloka, who explained, “The AI industry was for a long time heavily focused on the models and compute, and data has always been an overseen part of AI. Finally, [the industry] is accepting the importance of the data for training.” She emphasized that building models capable of chain-of-thought reasoning requires expert demonstrations to mirror the intricacies of human logic. Despite advancements in synthetic data generation, human expertise remains indispensable. Experts are crucial for ensuring accuracy, addressing biases, and maintaining trustworthy AI systems. Industry leaders increasingly view this “human-in-the-loop” approach as vital for achieving advanced reasoning beyond simple automation.