Inside the expert network training every frontier AI model | Garrett Lord (Handshake CEO)
August 24, 2025
AI Summary
5 min readHandshake CEO Garrett Lord explains how his 10-year-old career platform for college students and alumni—serving 20 million users, 1,500 colleges, and 1 million companies at $200 million ARR—launched an AI data labeling business in January. Leveraging its network of 500,000 PhDs and 3 million master's students, the new unit hit $50 million ARR in four months and is on track to exceed $100 million in its first year, working with seven frontier AI labs.
Post-Training Data for Frontier Models
AI model gains have shifted from pre-training on internet-scale data, which is asymptoting, to post-training for targeted improvements in areas like coding, math, law, finance, and STEM. Labs run experiments with data types including supervised fine-tuning (SFT: prompt-response pairs with step-by-step reasoning), RLHF (preference rankings), trajectories (screen recordings with narrated tool use), and rubrics (criteria for model-as-judge evaluation). Experts identify model weaknesses—e.g., via GPQA benchmarks where PhDs break reasoning steps or provide ground truth—then generate high-quality, novel data that generalists cannot produce, focusing on verifiable and non-verifiable domains.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
What you'll learn
- 1 **[05:07] Data Labeling Primer: Pre-Training vs. Post-Training**
- 2 **[09:48] Handshake's Expert Network Moat**
- 3 **[13:03] Real-World Data Tasks and Outputs**
- 4 **[19:52] Key Needs: Quality, Volume, Speed**
- 5 **[33:40] Launch Story: From Middlemen to 50M ARR**
- 6 **[45:48] Incubating New Biz Inside 10-Year Company**
- 7 **[53:50] Execution Tactics for Hypergrowth**
+ Full timestamped outline available in the app
Show Notes
Garrett Lord is co-founder and CEO of Handshake, which started as a career network for college students and new grads but recently discovered something extraordinary: they were sitting on the world’s largest network of academic experts—exactly what frontier AI labs desperately needed. With 500,000 PhDs and 3 million advanced degree holders creating training data, in just eight months they’ve built a new business that hit $50 million in revenue in its first four months and is on track to blow past $100M in the first 12 months.
What you’ll learn:
1. How Handshake found an opportunity to leverage their proprietary network of experts to launch a data-labeling business that’s on track to blow past $100 million ARR in 12 months
2. Why AI models need human experts (e.g. physics PhDs) to improve, and what this “data labeling” actually involves
3. Inside the actual work: what a biology PhD does for 8 hours that makes GPT-5 smarter
4. The playbook for building a startup inside a startup: separate teams, separate offices, separate everything
5. Why the shift from “generalist” to “expert” data labeling created a once-in-a-lifetime business opportunity
6. Why AI won’t eliminate entry-level jobs—it’s creating “Iron Man suits” that make junior employees 10x more productive
—
Brought to you by:
CodeRabbit—Cut code review time and bugs in half. Instantly: https://coderabbit.link/lenny
Orkes—The enterprise platform for reliable applications and agentic workflows: https://www.orkes.io/
Claude.ai—The AI for problem solvers and enterprise: http://claude.ai/
—
Transcript: https://www.lennysnewsletter.com/p/inside-handshake-garrett-lord
—
My biggest takeaways (for paid newsletter subscribers): https://www.lennysnewsletter.com/i/171410958/my-biggest-takeaways-from-this-conversation
—
Where to find Garrett Lord:
• X: https://x.com/garrettlord
• LinkedIn: https://www.linkedin.com/in/garrettlord/
• Email: [email protected]
—
Where to find Lenny:
< More from this podcast
Lenny's Podcast: Product | Career | Growth →