Technical writing on AI, engineering, and whatever I'm working through.
A deep dive into why context length exists in LLMs, what breaks when you exceed it, and the architectural constraints that make training long-context models expensive.
Exploring techniques for running large language models efficiently on memory-constrained devices, optimizing inference without sacrificing performance.
A comprehensive summary of the DINO paper exploring emerging properties in self-supervised vision transformers and their applications.
Deep dives on production AI — LLM inference, on-device models, and the systems around them. Roughly monthly.
Subscribe on Substack