FILE PHOTO: Figurines with computers and smartphones are seen in front of the words "Artificial Intelligence AI" in this illustration taken, February 19, 2024. REUTERS/Dado Ruvic/Illustration/File Photo
SAN FRANCISCO (Reuters) - - Artificial intelligence data startup Turing, one of a growing number of companies that provide human trainers to AI labs, said Tuesday its revenue tripled to $300 million last year as it reached profitability.
Palo Alto-based Turing, which says OpenAI, Google, Anthropic and Meta are its clients, was last valued at $1.1 billion in 2021.
As AI models have become more sophisticated, it has in turn increased demand for human trainers with specialized knowledge, boosting the valuation of startups like Turing competitor Scale AI, which was valued at $14 billion last year.
Based on what the AI companies want their models to get better at, AI data companies find workers with relevant expertise for those projects, reducing the burden of managing hundreds of trainers by the AI companies.
Turing says it has access to over 4 million human experts such as software developers or scientists with doctorate degrees, who it can contract to label data for AI models.
The costs aren't cheap: one complex annotation can cost hundreds of dollars, Turing said, and advanced AI models can require millions of annotations. For example, Meta used over 10 million human annotations when training the Llama 3 models, Meta executive Joe Spisak said last year.
As AI labs hit the "data wall," a term for when model performance plateaus due to lack of more internet training data, the labs will increasingly rely on human data companies to make their AI models smarter, Turing CEO Jonathan Siddharth told Reuters.
"Companies like Turing are helping the scaling laws keep going to make up for the data deficit that we have," he said.
(Reporting by Anna Tong in San Francisco; Editing by Kim Coghill)
