The Artificial PostAccount
All papers
Language models · ORIGINAL RESEARCH

A Training-Time Diagnostic for Generalization via the Log-Alignment Ratio

Selected research in language models. Read the full paper, including the methods, experiments, and reported results.

Opening the paper…

KEEP FOLLOWING THE IDEA

Meet the researchers.

Ashish Vaswani