Data 2.0: The Research Era of AI Data — Alex Ratner

Frontier Data Summit 2026: Alex Ratner explains Data 2.0, the shift from simple labels to complex environments for training and evaluating AI agents. At Snorkel AI’s Frontier Data Summit, co-founder and CEO Alex Ratner explains Data 2.0: the shift from simple labels to complex environments for training and evaluating AI agents. He covers task design, reward rubrics, quality control, human-AI data development, the evaluation gap, benchmark overfitting, and Snorkel AI’s $30 million Open Benchmarks Grants commitment. IN THIS VIDEO • How Data 2.0 differs from Data 1.0 • Tasks, environments, rubrics, and reward signals • Quality control and reward hacking • The human-AI data development loop • Open benchmarks, the evaluation gap, and benchmaxing SPEAKER Alex Ratner, CEO & Founder, Snorkel AI CHAPTERS 00:00 Welcome and summit overview 00:37 The $30 million Open Benchmarks Grants expansion 02:41 From Data 1.0 to Data 2.0 06:47 Tasks, environments, rubrics, and reward signals 11:01 Quality control and reward hacking 14:10 Finding useful data distributions 16:23 The human-AI data development loop 20:22 Open benchmarks and the evaluation gap 23:09 Benchmaxing and better benchmarks 24:57 Benchmark support and research fellowships RESOURCES Data 2.0 and the research era of AI data: https://snorkel.ai/blog/data-2-0-and-the-research-era-of-ai-data/ Open Benchmarks Grants: https://benchmarks.snorkel.ai/