There are automatic anonymization filters in DataClaw, and other solutions such as Pi's export. But my bet is that the concept of privacy is going to change as AI consumes everyone's data.
What gets me excited about this project is its legibility for philanthropy, like how the Internet Archive is the destination for random recordings found in Grandma's attic.
Having a trusted steward to handle the anonymization and eval design process could kickstart a world where retiring tax accountants, or even active and generous ones, are willing to donate their data to the posttraining commons at a scale that's actually much larger than the "data industry".
The real question is whether these security practices can truly survive the onslaught of 24/7 attack swarms, or if we're going to have to rebuild the entire cloud from the ground up.
It does feel like software and technology companies are undergoing a mass extinction, where the environment so negatively selects against certain traits that survival becomes impossible.
Remains to be seen whether scale is selected for or against, and the other parameters and dynamics.
I think we'll naturally develop intuition and infrastructure for fault-tolerant ways to implement LLMs in non-verifiable domains, i.e. requiring multiple, independent agents agree on a direction from multiple angles before taking an action.
This works well when the answer to the question is a number or something else definitive. When the answer is fuzzy, how do you compare 3 different responses? They'll all be different
reply