Hacker News Reader

Comment by deadbabe | Hacker News Reader

And then powerful local LLMs become feasible and everyone and every government self hosts and the megacorps lose all power.

Imagine the United States DoD, CIA, NSA training an LLM on all its top secret intel.

View on HN

I can’t imagine any agency in their right mind training an LLM on their top secret data… especially considering that essentially it would collapse all need to know data in one easily leakable single access domain. The models have no feasible or reasonable way to implement rbac. So you would end up in a situation where people who need to know who killed A, also know who killed B, not to mention cell A knowing potentially data on Cell B who is meant to watch them, etc, etc.

That’s the thing, the agencies just need to be run by people who are no longer in their right minds.

> it would collapse all need to know data in one easily leakable single access domain. The models have no feasible or reasonable way to implement rbac

Oh my sweet summer child...

This has been the continual goal of the US defense and intelligence agencies since 9/11, which was blamed on a lack of information sharing. Why do you think Edward Snowden had access to everything? Because it was consolidated post-2001.

You think there is any access control at the upper levels of the dark intelligence agencies? They have access to the whole take.

And it used to be that was pretty useless unless you had an actual lead. You'd need a East German Stasi level of labor to read everyone's secrets at scale... Now you don't, thanks to AI.

https://enwp.org/Total_Information_Awareness

Snowden had mass access to a mass surveillance program, not mass access to N dark programs or particular bits of high value intelligence. The thing you are describing is primarily around sharing threat intelligence or in other words, spying on us plebes. High value target intelligence and other bits of fun are still kept quite seperate.

The idea is silly because most of it is complete garbage anyway. You'd wind up with a model that might know a very small amount of actionable information, and the rest of it just speculative. Or extremely time sensitive, useful for only a relatively short slice of time (which has already passed).

A great idea that each person does the same. That makes it a draw, namely, a Nash equilibrium.

Reply on news.ycombinator.com