We tagged 340k threads in the Epstein archive by topic.
10k people, 500 aircraft, 400 banks, 100 government agencies.
Categories include Ghislaine's legal battles, career advice, dinner parties, and human trafficking.
Now available at jmail.world/taxonomy
Click on threads to open them in the classic Jmail interface, plus see other related topics that frequently co-occur.
We used GLiNER, an open source named entity extractor that gave us 920k noisy tags which we then filtered down to 36,000 distinct entities.
We then had cheap LLMs with knowledge of the Epstein case classify these tags and threads into a hand-crafted topic hierarchy.
This approach was inspired by Anthropic's post below
anthropic.com/news/how-peopl…
Another 403k email threads (Volume 9, 10) were added to Jmail thanks to our amazing team while this classification was happening, so this collection is incomplete. Expect it to improve and grow.
And most importantly, expect these tags to be embedded in the rest of Jmail as we clean them up!
Consider this a way to help you navigate, not an exhaustive list of every topic in these emails.
Use the feedback button to report mistakes, and expect these tags to get a lot better and more exhaustive over time.
The models we used are aware of the Epstein scandal, so you can dive very deep into pre-made collections like the one attached. Within the "Ghislaine Maxwell" category there's her managing staff, her managing the illegal operation's social presence, etc.
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.