Fine-tuning is mostly iteration: cleaning data beats tuning hyperparameters

ziv_ravid · x · 2026-11-20

Ziv Ravid's thread (posts 8-9): from there it's mostly iterations — train, look at where the model fails on the eval, figure out which kinds of traces are missing or noisy, fix, and repeat. Cleaning the data usually helps more than playing with hyperparameters, though it depends on the problem.

Related event: Engineer's Deep Dive Maps Enterprise Post-Training from Motive to Evaluation(22 posts)→

Original post →

More from coding & agent

coding & agent channel →