Hey there. Welcome to the third edition of this manmade newsletter. The goal here is to make it just a bit easier for all of us to understand what the hell is happening in the training data market. Who is buying? What data? Why?
Lab Diets
The training data value chain looks something like this:
VC money —> Lab —> Talent, Compute, and Data —> Model —> Revenue
As I wrote before, quality data is becoming the real bottleneck to hill climbing. Preferably, free range, wild-caught, organic data! Not factory farmed, inauthentic data. More on this difference in upcoming issue.
So, which lab is buying which data? And from whom?! Don’t be confused by the industry-jargon and flashy websites, it’s quite simple when you break it down…
Now you can just match the data types to the vendors. There are three good vendor lists circulating online. Sean’s discerning list, a crowdsourced list, and Chris’ original list that is somewhere in between.
New Landmen
The landman is a US invention from the 1860s that exists because in the US, private citizens own what's under their land. So when a valuable thing is found underground, companies negotiate farm by farm, deed by deed. That negotiator is the landman.
They started as “lease hounds,” speculators that got mineral rights from unsuspecting farmers and flipped them to brokers. Sound familiar? This is the Micro1 and Mercor referral army!
1859, Edwin Drake struck oil with his well inTitusville, Pennsylvania, prompting the first oil boom in US
These days, a landman has three jobs. First, run title (aka rights clearance in training data terms). Second, close deals. Third, scout for new geology / geography. This is exactly what the modern “dataman” is out there doing today!
There are millions of businesses around the world that have valuable organic, free range, wild-caught, organic data exhaust from their operations. The opportunity is to find them, understand the data use rights, and then convince them it’s worth it to “dig” or “build a pipeline.”
Some prescient investors are living in a future where new business are forming around data exhaust streams specifically to subsidize new business models or create new capital structures.
Quote of the Week
“We’ve developed the first robot foundation model which has learned how to learn new tasks instantly from only a single demonstration, whilst trained entirely in simulation without any real-world data”
Thanks for reading! You can reply directly to this newsletter with what I got wrong, what you want to see next week, or any other dank training data memes. See you next week 🫡