August Meeting - LLM Internals

This month we have Joe Smith returning to teach us about LLM Internals:

“It's Not Just Predicting the Next Word: LLM Internals for Curious Humans"

New models drop every week: 8B, 70B, 4-bit quantized, 128K context. What do those numbers actually mean? This talk is a picture-heavy, (mostly) math-free tour of LLM internals — what parameters and layers really are, why quantization fits giant models on your laptop, and what context windows and KV-caches do (and why long chats get slow). Along the way we'll bust a stubborn myth: that models "only think in language." What's actually inside — continuous representations where concepts blend and interpolate — is stranger and more interesting. You'll leave able to read a model card without squinting.

Agenda
6:00pm Network, Food
6:30pm Meeting start, announcements
6:40pm Last Month in AI
6:50pm Talk: Joe Smith