The $1B Al company training ChatGPT, Claude & Gemini on the path to responsible AGI | Edwin Chen
From Lenny's Podcast
Edwin Chen, founder and CEO of Surge AI, explains how his tiny, bootstrapped team built the data company behind frontier models like ChatGPT, Claude, and Gemini—hitting $1B in revenue in under four years by obsessing over “quality” rather than scale. The conversation digs into what it really means to “teach” models what’s good vs. bad, and why the values embedded in training data shape model behavior in ways most people overlook. Chen also argues that parts of the industry are steering AGI to...
Key Takeaways
- Skip the Silicon Valley playbook: tiny elite teams can outrun bloated org charts.
- In AI data, quality isn't headcount—it's craft; bodies don't magically make good labels.
- The values you bake into training data become the model’s morals, for better or worse.
- A helpful model should optimize your time, not trap you in endless “perfecting” loops.
- If we train for dopamine and slop, don’t be shocked when AGI acts like a tabloid.
Mentioned in This Episode
- rubrics (concept)
- Lenny (person)
- Terrence Rohan (person)
- Richard Sutton (person)
- Kevin Weil (person)
- Mike Krieger (person)
- Brian Armstrong (person)
- Albert Camus (person)
- Douglas Hofstadter (person)
- Noam Chomsky (person)
- Ted Chiang (person)
- Terence Tao (person)
- IDC (company)
- Vanta (company)
- WorkOS (company)
- Slack (company)
- Stripe Atlas (company)
- Cursor (company)
- Plaid (company)
- Koda (company)