'The era of data-labeling companies is over,' says the CEO of a $2.2 billion AI training firm

The landscape of AI training is undergoing a significant transformation, as traditional data-labeling methods lose their relevance. Jonathan Siddharth, the CEO of Turing, a prominent AI training company valued at $2.2 billion, declared on the "20VC" podcast that the age of data-labeling companies has come to an end. Siddharth emphasized that the requirements for data have evolved. Earlier AI models depended on manual tasks like tagging images and sorting text, which were highly scalable. However, current systems demand more sophisticated and real-world data that reflects genuine human knowledge work. He stated that top research labs are now seeking partnerships with AI training firms that can act as proactive research collaborators. According to Siddharth, the focus for AI training companies must shift towards creating reinforcement-learning environments—simulated mini-worlds that mimic human workflows across various sectors. Achieving this requires enlisting human experts from diverse fields. In June, Turing announced it had secured $111 million in Series E funding, bringing its valuation to $2.2 billion. The company also reported an impressive annual revenue run rate projected to reach $300 million in 2024, nearly tripling from the previous year. As AI training firms attract significant valuations, the sector is witnessing a rapid expansion of the freelance workforce. Reports indicate that many freelancers are earning substantial incomes from AI training tasks, although the work can often be unpredictable and challenging. Additionally, a shadow market for access to these platforms has emerged, with reports detailing numerous Facebook groups selling unauthorized access to both legitimate and fraudulent contractor accounts. While such practices are prohibited by AI training companies, the increasing demand for gigs in this field has created opportunities for scammers and opportunistic gig workers.