Search, in a browser?

Feed it to a local model?

If so: Wikileaks made/makes(?) all of their stuff easily browseable, "her emails" included.

You can search, but what do you search for if you don't know what's there?

If you don't know what's there you just start from the top I suppose. That's what happened with the Snowden leaks, they were picked apart over the span of months if not years by journalists, publishing what they found was interesting.

I suppose when you have a large corpus you start with as many browse points as you can. So maybe a list of the top senders, top recipients, recipient graphs.

I'd think by day would be next. Browse subjects by day, and word clouds for each message thread? The. Try to isolate topics discussed each day, then link the topics across days.

You should be able to click in and see the thread/messages/etc.

This is, at some level, how legal discovery stuff works from what I understand.

The part you're missing here is having a clue about what's sensitive in geopolitics in the first place, so you have that place to work from.

"Evil plans of new world order" obviously..

It seems you want a AI to analyze the data in general?

Otherwise you will have to do some work and read a little bit .. and then investigate to see if there is more. That is where the search tools are useful. Like in the example, finding out if that Scott guy changed his opinion on the 6. of january after Trump became president. (To see if the original statement was a lie. Not possible with ease so far)

People will speak in code, if they are planning crimes. Only some idiots speak openly of violent revolution in public messengers.

I feel like LLM agents for research across it could be interesting