Hacker News Reader

I really wish there was a large-scale corpus of completely benign email traffic.

500K messages from 150 users.

Now I’m just imagining a bunch of NLP models subtly learning large scale fraud.

AGI needs to be adept at fraud too, you know ;-)

Isn't it beautiful that there is a meaning of "benign" that does actually apply to these emails?

Amazon hosts a dump from the Apache email archives as of July 2011. IIRC, it contains about 7 million emails.