Field notes · 10 guides
Plain answers about a young market.
What labs buy, what your record is worth, how licenses work, and what terms like RL environments and trajectory data actually mean. Written for operators, not just researchers.
Selling data
- 9 min read
How to sell your company's data to AI labs
What labs actually buy, how to check you own it, and how a deal gets done without a procurement cycle measured in quarters.
- 8 min read
How much is your company's data worth to an AI lab?
There is no price list. There are, however, five things that move the number far more than size does.
- 7 min read
Licensing data vs. selling it: what you actually give up
The word "sell" scares people for good reason. Here's why almost every good deal is actually a license, and what to look for in one.
Training data
- 8 min read
What is operational data, and why do AI labs want yours?
The working record your business produces by accident turns out to be the scarcest input in AI right now.
- 9 min read
RL environments, explained for people who don't train models
Labs used to teach models by showing them answers. Now they teach agents by letting them practice. An RL environment is the practice field.
- 7 min read
Agent trajectory data: why labs want the messy middle
The answer is easy to find. The forty steps it took to get there almost never are. That's what trajectory data captures.
- 8 min read
RLHF and preference data, in plain English
How models learn taste, why "which answer is better?" turned out to be such a powerful question, and why the people answering it matter so much.
- 7 min read
Why private evaluation data is worth more than training data
Once a model has seen the test, the score stops telling you much. That is why labs pay so much for tests nobody has seen.