Archive — US — techcrunch
Why AI startups are taking data into their own hands
AI startups are increasingly taking data collection into their own hands, hiring freelancers to create high-quality datasets for training models. Turing Labs uses human-drawn video data for its vision models, while Fyxer highlights quality over quantity in data preparation. This shift emphasizes proprietary, curated datasets as a competitive advantage in AI development.
Key points
- Turing Labs employs freelancers, like Taylor, to create high-quality video data for AI training.
- The work includes capturing diverse manual tasks to improve the AI's visual reasoning skills.
- Turing's Chief AI Officer emphasizes the importance of varied datasets for effective model training.
- Fyxer, another AI company, focuses on tailored data with a strong emphasis on quality over quantity.
- Both companies believe that effective data collection is a critical competitive advantage in AI, leading to more effective models.
Why it matters
- Growing trend in AI to prioritize proprietary training data reflects a shift in data sourcing practices.
- Historical reliance on scraping public data and low-cost labor is evolving as companies realize the value of curated datasets.
- The need for high-quality data aligns with the 2025 IT trend of data privacy and ownership, with firms looking to protect their data assets.
The shift towards in-house data collection and high-quality training datasets marks a pivotal evolution in AI development, illustrating how startups are recognizing the importance of data as a key competitive resource.