Lenny's Podcast · December 7, 2025

The $1B Al company training ChatGPT, Claude & Gemini on the path to responsible

with Edwin Chen

Edwin Chen reveals how AI training is fundamentally broken, with models being optimized for engagement over truth through flawed benchmarks and leaderboards. Learn why LLM Arena rankings prioritize superficial qualities like emojis over accuracy, how quality training data resembles Nobel Prize-winning poetry, and why academic benchmarks mislead AI development. Discover the counterintuitive insight that small elite teams outperform large tech organizations, and how AI models will differentiate based on company values rather than becoming commoditized. Essential listening for understanding the future of responsible AI development and avoiding common optimization traps.

Featured insight

We're building AI that teaches models to chase dopamine instead of truth, optimizing for engagement rather than advancing humanity through curing cancer, solving poverty, and understanding the universe. — Edwin Chen

Best for: AI/ML Engineers building production systems, Startup Founders developing AI products, Product Leaders at AI-first companies

Loading the full episode…