I think of it as, we've had a hurdle where computers were too slow to organize the data humans produced, but have since crossed it. LLMs and CV also enable analysis of fuzzy data, which hasn't been possible before. This makes me uneasy, to say the least.
Precisely, modern multimodal LLMs are a breakthrough in ultra high volume data processing, for all their flaws
I asked Claude to analyze and datamine all of my eleven thousand HN comments, plus my reddit comments, lobsters comments, mailing list emails, stackexchange content, etc. With this database it managed to create an entire dossier about me. It could psychologically profile me, and it had a better and more complete view of me than my literal psychologist. We identified things like weaknesses, social engineering susceptibility, likely scenarios. It got to the point it started digging into open source projects I made when I was a teenager and started making eerily accurate inferences about my childhood.
I can't even imagine what the NSAs and CIAs of the world are capable of with their limitless black budgets. Anthropic only balked at domestic surveillance, it had zero objections towards allowing the US government to surveil foreign persons like me.