535: The You in the Machine
Oct 1, 2026 · 1h 29m
More Technology podcasts · More Science podcasts
Summary
Professor Adam Smith joins the show to discuss the rapid evolution of small language models (SLMs) that run locally on phones and laptops. The conversation covers the shift from massive frontier models to efficient, open-weight alternatives, explaining how context distillation and tool-calling capabilities allow these smaller models to handle complex tasks. Smith also explores the concept of "trans agency," using LLMs to automate routine workflows while maintaining human oversight, and offers practical advice on leveraging local models for offline use and personalized automation.
Topics discussed
Introduction and guest welcome
Lightning round: preferences and quick takes
The rapid pace of LLM development and local models
Scaling trends: frontier vs. local model capabilities
Using LLMs for automation and shell scripts
Fine-tuning small models with context distillation
Defining tokens and the economics of AI usage
Bayleaf: A campus-scale open-source AI service
Open-source tools and privacy-focused local setups
Inference providers, cloud infrastructure, and hardware
Model size, efficiency, and the 'vehicle' metaphor
Recreational LLM use and skill acquisition
Community resources and advice for ADHD users
Closing thoughts: notes, identity, and human-machine collaboration
More from Embedded
Listen ad-free on Castria