Context is intelligence.
A language model’s capabilities grow with compute, and there are two ways to spend it: parameters or context.
Fern’s architecture, training and serving support 10M tokens (~23 MB of code) at once, and it uses them: prediction keeps improving up to 10M.