Turing’s New Initiative: Training AI Using Everyday Work Videos

Staff Reporter | PNN:
U.S.-based company Turing is pioneering a novel approach to training artificial intelligence (AI). The company is collaborating with artists, construction workers, electricians, and even cooks, capturing videos of their daily tasks to train a new Vision Model.

For a week, artist Taylor and their partner wore GoPro cameras on their heads while painting, sculpting, and doing household chores. These videos are being used to train Turing’s AI model. The goal is for the AI not only to recognize objects visually but also to understand real-world workflows and step-by-step problem-solving processes.

Taylor said, “We would start our daily tasks in the morning, then turn on the camera for cooking, cleaning, and art-making. We had to provide five hours of footage each day, but actually worked seven hours because it was physically exhausting.”

Turing’s Chief AGI Officer, Sudarshan Shivaraman, explained that their goal is to record diverse hand-based work so that the model can learn how different tasks are performed across professions. “We’re selecting people who work with their hands—cooks, masons, mechanics, artists. This diverse data will be the strength of future vision models,” he said.

Currently, 75–80% of Turing’s data is synthetic, meaning it is artificially generated from real video. However, the company emphasizes that the higher the quality of the original footage, the more effective the synthetic data becomes.

This new approach marks a major shift in AI research—where previously models were trained by scraping countless images and videos from the internet, companies are now producing human-centered datasets themselves.

Turing is not alone. Another AI company, Fyxer, is taking a similar approach. Founder Richard Hollingsworth explained that in training their AI system for email organization and responses, “we realized it’s not the quantity of data, but its quality that matters. Good data equals good performance.”

He added that their model training involved experienced executive assistants because “human experience is essential to understanding the nuances of email context and interpersonal communication.”

Analysts suggest that Turing and Fyxer’s strategy is creating a new trend in AI development—where the true competitive advantage is not the algorithm but high-quality, human-generated data. As Hollingsworth puts it, “Anyone can use an open-source model, but not everyone can find skilled people to create the data—this is our real strength.”