[tl;dr sec] #331 - How Adversaries Use AI, Skill Issues, Using IDEs for C2
Brief
Hey there,
I hope you’ve been doing well!
👩❤️👨 Repo-mantic Comedy
Recently I had one of those moments where you remember that LLMs are trained on the vast, beautiful, complicated collection of human knowledge.
I was using Codex to port a feature from one code base to another, and it said:
“…I’m reading the exact code paths now so the port preserves behavior instead of inventing a prettier cousin .” 😂
Dear reader, I had questions. Like: how many bodice ripper novels and country music lyrics are in the training corpus? What other secrets lie in the weights?
I didn’t think I read anything about dating preferences in its model card.