Edwin Chen, founder of Surge AI, on what frontier labs actually pay for

Edwin Chen, founder and CEO of Surge AI, explains why he thinks today's frontier models are so uneven. He points to one benchmark where AI went from solving just 5% of problems written by Ivy League professors to cracking century-old math, and another where the same models still fail more than 70% of the time at ordinary office work. Edwin believes closing that gap is where AI's real progress may happen next. Watch more ICONIQ Frontier conversations: https://youtube.com/playlist?list=PLLJ7v4jTuYf0&si=uAhE-x3lJ_7gNWAa Key Moments: 02:18 Why Surge skipped computer vision's cat-or-dog problems 04:03 The infinite ceiling of quality in frontier models 05:27 Fremont Bench, and models that now solve century-old math 06:21 handbook.md, and why models still fail an ordinary workday 08:04 Inside an RL environment, a startup built for models to work in 10:15 Real company data versus synthetic data, and why Surge uses both 11:39 Why Surge chose quiet, quality-first growth over hype 13:31 Surge's benchmarks landing on the Claude model card 14:28 The post-training link between data quality and cybersecurity 17:00 Why Surge has stayed lean relative to the business it built 18:43 How Edwin's own curiosity shapes what Surge builds Stay connected: Follow us on LinkedIn: https://bit.ly/3V8wH0b Follow us on X: https://bit.ly/4874SwQ Explore ICONIQ Insights: https://bit.ly/45Vj22R Disclaimer: iconiqcapital.com/outreach-disclaimer