20VC: The Biggest AI Leaders on What Matters More; Model Size or Data Size & Where Does The Value in AI Accrue; to Startups or to Incumbents

Data Moats

The concept of data moats is pivotal in determining the competitive edge in AI. Douwe Kiela highlights that while incumbents have access to vast proprietary data, much of the valuable data is freely available online, allowing startups to train high-quality models without exclusive datasets 1. Alex Lebrun adds that incumbents benefit from distribution advantages but often lack the agility to innovate radically, potentially leaving room for startups to disrupt with new paradigms 2. Richard Socher emphasizes the dual nature of data access, noting that while unsupervised data is abundant, proprietary data remains a significant advantage for incumbents in specific applications 3.

If you can train on that language but nobody else has it, that puts you in a position of advantage.

--- Douwe Kiela

This dynamic creates a complex landscape where both startups and incumbents can thrive, depending on their data strategies.

   

Proprietary Data

Proprietary datasets are crucial for AI startups, influencing venture capital decisions and strategic directions. Douwe Kiela explains that having a robust data flywheel can serve as a moat for deep tech AI startups, enabling them to generate and leverage large datasets for competitive advantage 4. However, he also notes that large language models are increasingly data-efficient, allowing startups to innovate with minimal data, which was previously unfeasible 4. This efficiency opens up new possibilities for startups to create specialized models without the need for extensive proprietary data.

You want to start with a lot of data and then have a way to generate lots more data, and that data is going to be your moat.

--- Douwe Kiela

Thus, while proprietary data remains important, the landscape is shifting towards more accessible and efficient data utilization.