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

""Data needs have significantly changed," said Siddharth. Early models relied on annotators tagging images, classifying text, or performing simple tasks that could be outsourced at scale. Today's systems, like agentic models and reinforcement-learning architectures, require more complex data, he added. "It's more real-world data, data that touches how real humans do knowledge work," Siddharth said, adding that major labs want to work with AI training companies that can be a "proactive research partner for them.""

"Siddharth said AI training companies need to focus on building a reinforcement-learning environment - simulated mini-worlds - that replicate human workflows across different industries. To do that, AI training companies must recruit human experts in various domains, Siddharth said. Turing announced in June that it had raised $111 million in Series E funding at a valuation of $2.2 billion. Earlier this year, the AI training firm said its annual revenue run rate reached $300 million in 2024 - nearly triple the year before."

Basic data-labeling work built on tagging images or sorting text is becoming obsolete as AI models demand more complex training data. Modern systems, including agentic models and reinforcement-learning architectures, require real-world data that mirrors how humans perform knowledge work. AI training must move toward building reinforcement-learning environments—simulated mini-worlds—that replicate industry workflows and incorporate domain experts. Major labs prefer training partners that act as proactive research collaborators capable of designing such environments. AI training firms have seen rapid growth and large valuations, exemplified by Turing's $2.2 billion valuation, recent funding, and a substantially increased revenue run rate.

Read at Business Insider