@profdiggity@privacysafe.social
2026-09-21 17:04 UTC
@wbai@bird.makeup In fact, I've been warning about the dismantling of guardrails since the rise of so-called "cloud computing" and the constant leaking of sensitive data across networks. Generative or hybrid AI systems add fuel to this fire because they are mostly probabilistic, and humans can have difficulty communicating their goals or intentions to them. In the machine learning literature that's broadly called "the alignment problem."
Replies (1)
-
@profdiggity@privacysafe.social 2026-09-21 17:04
@wbai@bird.makeup But it's also a familiar problem for management of any system with complexity, as military and economic history warns us. I don't see LLMs as being uncontrollable, at least not in some way where we should think about them as distinct and different from other software. Where #infosec is concerned, the problems we're facing are mainly the increasing scale and frequency of attacks coupled with decreasing knowledge and effort on the part of human operators.