S3 E3 Scale Founder CEO Alexandr Wang: in AI your data writes the program

In this talk

1. 4:15

   From data labeling to applied AI

   Wang describes Scale’s two businesses: Scale Data helps AI labs build high-quality datasets, while Scale Applied AI provides models and products for business use cases. The company expanded into applied AI after seeing that only a small share of the economy could afford to build cutting-edge models.

2. 10:04

   Why data became the starting point

   While building an image recognition model for a camera in his fridge, Wang spent substantial effort collecting and labeling tens of thousands of food images. He concluded that deep learning code often stays mostly fixed while data varies by application, making data a way to shape how a model behaves. He says Scale aimed to make this overlooked work useful and dependable.

3. 14:44

   Operations and quality as an advantage

   Wang argues that combining technology with operations can create a durable business advantage, drawing parallels with Amazon’s logistics and cloud operations. At Scale, machine learning can speed up tasks such as outlining objects, while human review and machine checks can catch different errors. He also describes three points where information can be lost: writing labeling guidelines, applying them, and training the model.

4. 41:05

   Foundation models and human feedback

   Wang says foundation models have made AI more accessible, though reliability remains a difficult final step. For language models, he sees human feedback as important for usefulness, truthfulness, evidence, and alignment with user intent. He describes RLHF guidelines as design choices that shape how people rank responses, and says this work has become a growing part of Scale’s business.

5. 30:20

   AI in national security and industry

   Wang explains his interest in national security through his upbringing in Los Alamos and his parents’ work as physicists. He gives an example of using satellite imagery models to assess building damage in Ukrainian cities for humanitarian and conflict response. He also expects AI to reshape areas such as advertising and health care, including automated ad variants and support for medical diagnosis.