Flutterby™! : A Gentle Crash Course to LLMs

Blaise Brignac writing on Specter Ops: A Gentle Crash Course to LLMs, particularly for its long discussion of security issues.

As previously discussed, LLMs are just brains in a jar operating in much the same way a hyperintelligent 4-yo would after binging on state fair sweet tea and cotton candy. To correct this, they have been wrapped in agentic structures, so we need to talk about that.

There's obviously a lot of stuff with having the LLMs write prompts to have less privileged LLMs do subtasks, and work through layers of that, and this discusses some of those mitigation strategies, but... yeah... this is more "let's give random things access to our data" with levels of obfuscation that package management repos can only dream of...