Entries, Model Internals | Paula Livingstone

Projects

Model Internals

A hands-on, evidence-first investigation into what actually happens inside a running language model, opening one up on local hardware and testing directly why it does what it does.

Teaching a Model to Use Tools

1 entry

Where a Model's Uncertainty Actually Lives Give a model two real tools and a genuinely ambiguous question, and the big decision, which tool to call, is made without a flicker of doubt. Its real uncertainty turns out to live somewhere far more human.