Generalists to Experts Shift

As models improve, the value shifts from generalist labor to domain experts who can break the models

Garrett Lord
Inside the expert network training every frontier AI model

Generalists to Experts Shift

"The models have gotten so good that the generalists are no longer needed. What they really need is experts, experts across every area that the models are focused on." - Garrett Lord

What It Is

The Generalists to Experts Shift describes a fundamental market transition in AI data work. Early AI training relied heavily on generalist labor—people in low-cost markets drawing bounding boxes around stop signs or doing basic classification tasks. As models have improved, this generalist work has diminishing value. The frontier has moved to expert knowledge that can actually challenge and improve advanced models.

This shift represents a broader pattern: as AI capabilities advance, the human contribution that remains valuable becomes more specialized, not less. Being "smart" or "capable" isn't enough—you need to be smart in ways that models are not yet smart.

How It Works

The Old Model (Generalist Data):

  • Basic tasks: labeling images, classifying text, simple preference ranking
  • Labor source: lower-cost international workers, anyone who could follow instructions
  • Value driver: cost efficiency at scale
  • Moat: operational scale and logistics

The New Model (Expert Data):

  • Advanced tasks: breaking models in specialized domains, providing ground truth in physics/chemistry/medicine, correcting reasoning chains
  • Labor source: PhDs, domain experts, people who've extended human knowledge
  • Value driver: quality and expertise that models can't replicate
  • Moat: access to experts and their trust

Why the Shift Happened:

  • Pre-training (consuming all internet data) hit diminishing returns
  • Post-training gains now come from expert-quality data
  • Models are now good enough that average people can't break them
  • The remaining improvement requires knowledge that isn't on the internet yet

How to Apply It

  1. Recognize where expertise creates value - In your market, identify where deep domain knowledge is becoming more valuable as automation improves baseline tasks.

  2. Reposition for expertise - If you've been competing on generalist capabilities, look for the expert angle that will matter as the market shifts.

  3. Build expert networks - Access to genuine experts becomes the scarce resource. Start building relationships before you need them.

  4. Update your talent model - Stop optimizing for low-cost labor; start optimizing for credentialed expertise that commands premium rates.

When to Use It

  • When evaluating market positioning as AI capabilities advance
  • When deciding what human capabilities to invest in
  • When the work that made you successful is becoming commoditized
  • When considering which expert communities to build relationships with

Example

Garrett Lord describes how Handshake's positioning evolved:

Before the shift: Companies recruited generalist annotators through mass advertising, paying modest rates for basic labeling work. Success meant operational efficiency.

After the shift: Frontier labs need physics PhDs who can prove where GPT-4 is wrong in thermodynamics. They need chemists who can provide ground truth in organic synthesis. The average person literally cannot break these models—only experts can.

Result: Handshake's network of 500,000 PhDs and 3 million master's students became extraordinarily valuable overnight. Their decade of building relationships with students at top STEM programs created competitive advantage they didn't originally anticipate.

Source

  • Guest: Garrett Lord
  • Episode: "Inside the expert network training every frontier AI model"
  • Key Discussion: (00:10:50) - Explaining the market shift from generalists to experts in AI data
  • YouTube: Watch on YouTube

Related Frameworks

Back to Growth