Why I Started Stokes AI: A Personal Story

Over the past few years, I’ve spent countless hours working with some of the leading companies in AI data labeling and evaluation — specifically, DataAnnotation.tech (Surge AI) and Outlier.ai (subsidiary Scale AI). Through these roles, I contributed to the training and fine-tuning of large language models by ranking model responses, evaluating factual correctness, writing supervised completions, and more.

It was rewarding work. I saw firsthand how the quality of human feedback can shape the behavior of powerful models. But over time, I started to feel like I could do more — not just as a single contributor, but by building a team and a company that focuses on precision, domain expertise, and flexibility in AI training.

That’s why I created Stokes AI.

The goal is simple: to push the frontier of human-in-the-loop AI training. I wanted to build an organization that’s nimble enough to take on customized projects and rigorous enough to deliver high-quality, expert-labeled data across a wide range of tasks — from preference ranking to adversarial prompting to domain-specific evaluation.

Generative AI models are only as strong as the data they learn from. And as the industry moves forward, I believe it’s critical that model developers have access to diverse, independent, and deeply thoughtful human supervision — especially in high-stakes domains like STEM, education, finance, and policy.

Stokes AI was born from that belief.

Whether you're an AI research lab, a startup fine-tuning a model, or just someone curious about the future of AI training — I appreciate you being here. This is just the beginning, and I’m excited to see where it leads.

— Anderson Lu
Founder & CEO, Stokes AI