This has to be one of hte most fascinating discussions i've seen on HN. Imagine the FBI training an AI on all the information they have about a suspect. Files and files of statements, social media history, phone taps, etc and then interrogating the AI to get enough information to convince a judge to issue a warrant.
idk the LLMs being turned loose and the possibilities feels different than past major shifts in tech. (i've been around a while)
That seems less efficient than just searching through the contents without training an LLM to adopt the persona of the suspect. You'd have to double check everything they say, because obviously any of it can be false. So at best it's pointing you to the right direction, but realistically not.
Might be a good lead generation if the FBI is completely stumped I guess? Or they're just bored at work?
idk the LLMs being turned loose and the possibilities feels different than past major shifts in tech. (i've been around a while)