Lenny's Podcast: Product | Career | Growth

Inside the expert network training every frontier AI model | Garrett Lord (Handshake CEO)

958 snips
Aug 24, 2025
Garrett Lord, co-founder and CEO of Handshake, shares how his career network evolved into a powerful data-labeling business tapping into a vast network of academic experts. He discusses the crucial role of expert data in training AI models, illustrating how PhDs help make models like GPT-5 smarter. The transformation from generalist to specialist labeling has opened new business opportunities. Garrett also touches on the collaboration with AI labs, and how AI technology enhances productivity, likening it to giving junior employees 'Iron Man suits.'
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes

Fix Models By Providing Stepwise Ground Truth

  • Break models by finding where reasoning or steps fail, then provide ground truth and step-by-step corrections.
  • Use expert-created trajectories, rubrics, and SFT data to teach correct intermediate steps, not just final answers.

Quality, Volume, Speed — Lab Priorities

  • Labs care first about data quality, then volume, then speed when choosing labeling partners.
  • Handshake built tech, post-training teams, and rented GPUs to evaluate each unit's likely model impact.

Treat Expert Contributors With Respect

  • Treat experts like experts: pay well, train them, and integrate their learning back into their careers.
  • Enable fellows to earn meaningful hourly rates and bring insights back to academia and research.
Get the Snipd Podcast app to discover more snips from this episode
Get the app