From 17% to 97.8%: Making a Laptop-Sized AI Actually Reliable
Play episode6:10
About this episode
A laptop-sized model went from 17% to 97.8% on a real task. Sam and Maya on why the prompt, not the model, was the biggest lever.
In this episode
- Why the prompt was the problem
- Fine-tuning as polish, not the fix
- Making it fast enough to use
Sam and Maya unpack the field note From 17% to 97.8%: Making a Laptop-Sized AI Actually Reliable: the argument, the numbers and the practical moves, in a short, sharp conversation.
The Growth Project