Reactions from ranked influencers
2 posts
Surgical Repair of Insecure Code Generation in LLMs Neat paper from XBOW. Building models that generate more secure code by identifying failure categories and addressing them one by one. I appreciate work that gets into the data, determines themes, and addresses them individually; that's how you understand, and build lasting fixes. https://arxiv.org/abs/2604.16697
@LeonDerczynski I appreciate the shout-out, but all credit goes to @gussand and @sg1753 back at NYU! :)