The $1B Al company training ChatGPT, Claude & Gemini on the path to responsible AGI | Edwin Chen
Edwin Chen of Surge AI explains how $1B in revenue with <100 employees was built on data quality, not VC hype, and why AI benchmarks are optimizing for slop.
- Surge hit $1B+ revenue in under 4 years with fewer than 100 employees, fully bootstrapped and profitable from day one — possibly fastest ever.
- Chen attributes Claude’s coding/writing dominance to Anthropic’s taste and willingness to optimize for real-world performance over benchmark PR.
- LM Arena and most AI benchmarks are gamed: adding emojis and doubling response length reliably climbs leaderboards regardless of accuracy.
- Models can win IMO gold medals but still fail at parsing PDFs — benchmark difficulty has no correlation with real-world usefulness.
- AGI is a decade or more away; automating 80% of an L6 engineer’s job takes 1-2 years, but 99% automation takes much longer due to exponential difficulty curve.
- AI models will diverge sharply by company values, not just capability — the objective function each lab chooses will define model personality and behavior.
- Vibe coding is overhyped: dumping AI-generated code into codebases creates long-term maintainability debt most people aren’t accounting for.
- Chen founded Surge one month after GPT-3 launched in 2020, after repeatedly hitting data-quality walls at Google, Facebook, and Twitter.
2025-12-07 · Watch on YouTube
