Edwin Chen

Edwin Chen is co-founder and CEO of Surge AI — the fastest company to reach $1B in annual revenue (four years, under 100 people, bootstrapped). Surge trains and evaluates frontier AI models for all major AI labs. Former researcher at Google, Facebook, and Twitter. Background in mathematics, computer science, and linguistics (MIT). Built Surge one month after GPT-3 launched in 2020.

Key ideas

• Quality is not checkbox-completion. Real data quality requires implicit, subjective, complex assessment — Nobel Prize-winning poetry, not poem-with-eight-lines. Surge builds ML systems with thousands of signals to identify genuine quality, not surface compliance.
• Benchmarks are broken. Two failure modes: wrong ground-truth answers; and objective-answer structure that enables hill-climbing disconnected from real-world performance.
• AI slop thesis. Labs optimising for engagement metrics (LLM Arena, sycophancy, time-on-platform) are building models that chase dopamine instead of truth. Wrong objective function → wrong AI.
• RL environments are the next frontier. Full simulation of real-world messy contexts; long-horizon multi-step tasks; trajectory evaluation matters as much as final-answer correctness.
• Model values differentiate models. Lab values shape post-training choices. Models will become more differentiated, not commoditised — reflecting the ethical and aesthetic choices of their builders.

Appearances

See also

• Reinforcement Learning from Human Feedback
• Fine-Tuning
• Evals
• Sycophancy