Blog

How Transformers Learn Where Words Are: A Visual Guide to Positional Encodings

October 13, 2026  ·  transformers positional-encoding LLMs tutorial

Imagine handing someone three cards: dog, bites, and man.

Read more →
Why LLMs Are Numerically Fragile: A Researcher's Perspective

May 17, 2025  ·  research LLMs robustness

One of the more surprising findings from my recent research is how numerically fragile large language models actually are. We’re used to thinking of these systems as remarkably capable — able to reason across domains, write code, and engage in...

Read more →