Bruno Sanchez is the Executive Director of the Clay Foundation, which just released their v1 of a “Foundational Model of Earth”. We talk about what that means, building open source & non profits and can’t help but draw parallels to the not-so-open anymore OpenAI. Previously Bruno was the Program Director of the Microsoft Planetary Computer

Sponsor: OpenCage

Use OpenCage for your geocoding needs with their API
Geomob

Note: Links to books are Amazon Affiliate links. I earn a small commission if you buy any of these books.

• Clay
  • Github repo
  • Website
• My previous interview with Bruno
• Book & Podcast recommendation
  • A Thousand Brains by Jeff Hawkins (Affiliate Link)
  • Bad Therapy by Abigael Shrier (Affiliate Link)

(00:00) - Intro

(01:59) - Sponsor: OpenCage

(03:39) - What is a Foundational Model?

(08:45) - Foundational Task

(11:00) - Embeddings, and why they matter

(13:49) - Comparing to compression algorithms

(16:57) - What do embeddings enable?

(19:50) - Finding the Relationship between data

(24:16) - Implementation of monitoring all the data - I believe this is where you're going with the question

(25:44) - Implementation details

(27:52) - Validating a foundational model

(33:49) - Earth is a "limited problem"

(42:47) - Funding as a non-profit

(47:53) - Raising enough philanthropic funding

(53:02) - A Litmus test for open projects

(56:26) - Future predictions

(58:01) - Transparency/privacy

(01:00:53) - Commercial data

(01:03:36) - Openstreetmap

(01:05:49) - Updating foundational models

(01:12:17) - Clay model v2?

(01:13:43) - Dataset to embedding size comparison

(01:14:41) - Model v2 - continued

(01:16:45) - Difference working at a non-profit vs a corp

(01:20:27) - Book & Podcast recommendation

(01:24:01) - Support the podcast on Patreon!

• Support the podcast on Patreon
• My Twitter
• Podcast Twitter
• Read Previous Issues of the Newsletter
• Edited by Peter Xiong
• Find more of his work