Field Notes · Closing

I accidentally turned LLM memory into program analysis

This article describes how a technique intended for storing context in language models was repurposed to analyze program behavior and properties.

Show HN: Talos – An AI agent with a permission kernel between model and shell

Talos is an AI agent that implements a permission system between a language model and shell access to control what actions the model can execute.

Investigation of agents' behavior in the OpenAI/HuggingFace hacking incident

This item investigates how agents behaved during a security incident involving OpenAI and HuggingFace systems.