A written account of my HAICON 2026 talk on agentic AI: the shift from asking questions to delegating responsibilities, and the open challenges ahead.
From a 1994 compression algorithm to the tokenizer powering modern LLMs: how Byte-Pair Encoding works and why it became the standard.
Working through the edge cases of vocabulary design to understand why modern LLMs use subword tokenization.
Interactive tools and visualizations for teaching LLM architecture, gathered from preparing a workshop on LLMs.