Consume Product

The problem is that LLM/AI is unreliable. It can go wrong and do stupid things, even if most times it's fine. In video games AI is common, and I do not know any game where you cannot exploit it somehow. In various RTS there simply are strategies and units it cannot handle and goes full retard. In a game like World of Warcraft, where you have pathfinding, you can walk up and down some stairs and cause the NPCs to walk around the room back and forth.

Under normal circumstances AIs do work. And even if sloppy, it's usually good enough. But when it comes to arresting people, it is not reliable at all - it is a serious matter that needs actual people to investigate. At best an AI can only point towards potential cases.

In my case, the LLM *can* be helpful, but I have to dedicate quite some effort to ensure that my query is useful for work. Otherwise it's worthless slop that doesn't help at all - or I'd have to converse back-and-forth for ages until it finally gives me something useful, which is a frustrating and exhausting process.

In the hands of idiots, AI is dangerous only because it empowers idiots to do idiotic things. Not because AI does anything on its own. It just elevates stupidity into dangerous idiocy. In the hand of cops... it's the worst case possible.