About
We started Between Tokens because we had too many questions.
We wanted to understand how AI is actually built, not just read the announcement.
We’re Raghav Sharma and Manan Mehta.
What happens inside an AI lab? What does a billion-dollar training run actually look like? How does a model learn a new skill after pretraining? Can an AI researcher help build a better version of itself? Can AMD or Intel loosen CUDA’s grip?
We kept finding short answers to long questions. So we decided to work through them ourselves and share what we learn.
One question at a time§
For each investigation, we read what is public, make assumptions where the record goes quiet, run the numbers, and trace the decisions from the model to the business around it.
We won’t always get everything right. We show our work so other people can question it, correct it, and build on it.
One large question. Every consequential decision.
Working on an AI problem?§
We also set aside a few free, half-hour sessions for teams working through a real AI problem.
Send us the messy version: the workflow that wastes time, the product idea you can’t quite shape, or the model that keeps failing. We’ll help you find one useful next step.