Jmail maintainer and co-creator here. Very excited to see that someone finally made Jemini good!
Our development process has been interesting. Although just Riley and I first made Jmail, it's been really gratifying to see companies, journalists, and fellow developers like Diego rise to the occasion to make this entire suite of apps as high quality and extensive as possible.
Great, I've been looking to be able to contact you somewhere, hardly a better place :) Thanks for sharing the project btw, and for keeping jmail alive, been useful to dig into some stuff.
For that example, the first problem is that there is no link to the source file from the disclosed files, which is strange, because most other emails have that. Secondly, this almost seems to be a "sponsored" email or something, as it has an ad in the top right corner reading "Sponsored by Drop Site News", but clicking that just takes you to some general page, with no clear information how that's related to the fake/injected email. Also, what does "Verified by X" actually mean, did they verify the authenticity of that specific email?
There seems to be a bunch of people confused by those emails and wondering where they are coming from, because it's missing the source link like the others, so there doesn't seem to be a good way to verify those emails. Could you maybe share a bit about what's going on?
Has a "View Original Document" which takes you to kind-of-the-source (ideally should have been link to justice.gov but better than nothing I suppose), but for the emails with the Dropsite ad, it seems there is no original document?
Many Yahoo emails do show you the original source, and the original is just an EML file. These were files directly exported from Epstein's Yahoo account. Bloomberg used these EMLs to confirm that the Yahoo emails are real (https://www.bloomberg.com/news/newsletters/2025-09-12/epstei...).
We don't tamper with these EMLs, so we currently don't make the EML accessible if the team had to redact any contents of that email.
Also, that says "Verified by Drop Site News", not Sponsored by. That's because Drop Site redacted these real Yahoo emails and gave them to Jmail. The original Yahoo dataset, which the DOJ and House Oversight Committee did not release, is stewarded by DDoSecrets (https://ddosecrets.org/article/epstein-emails).
That’s just a campaign email with “Pedophiles” as the recipient name, right? Anyone can sign an email address up to a mailing list. All of Trump’s campaign emails are these type of overly friendly weird junk.
Sure, but the email itself doesn't seem to be a part of the archives of Epstein's emails (which would be an issue), it seems like it has been added manually by the admins of Jmail, as it's not coming from the files that were recently released.
“The Jmail Suite is an interactive archive of Jeffrey Epstein's emails, documents, photos, and more. Data compiled from the House Oversight Committee, Department of Justice, and DDoSecrets releases”
If I understand correctly, the emails you’re looking at are from a leak that is only accessible to journalists — so Drop Site News (as journalists) have access and have published some.
The Jamazon interface is really nice. I like how clicking on the order number takes you to the email. But I'm not going spend much time on there, after seeing his order for Lolita next to orders for kid toys.
I wonder if we can trust Gemini to do its job well here? Whoever is being protected in those files obviously has the power to compel governments to do what they want - if Gemini started being a threat, I bet it would get some "alignment" help. Certainly its findings would be reported, as well as the identity of whoever was doing the prompting
Interesting stuff! I'm also slowly diving into the whole Epstein thing, mainly focused on a human-curated wiki (at https://corroborators.wiki, still WIP obviously) and I've had some contributors wanting to use AI for basically filling out details automatically. But I'm using LLMs daily for programming, so I know how much they get wrong.
So about that specific thing, how do you avoid this particular problems with LLMs getting basic things wrong and basically being a lottery if the answer is correct or not? It's the reason I've avoided (and encouraged others to avoid it to) using LLMs too heavily in the process of reading, understanding and summarizing documents, but clearly you've must have figured out some trick to this?
Why on earth does he have CPAP pipes, an oxygen concentrator, and a pulse oximeter? This is dated 2019 so I assume he was having sleep apnea, but in my nightmares I assume he's keeping someone in a coma.
Great! This, along with Webb, the other Epstein AI tool, are amazing effort to find documents. The way you link to files instead of just outputing an answer is amazing. Great works!
When LLMs first came out, my initial impulse, though I didn't pursue it, was that the summarization function was ideal for sorting through FOIA requests. So glad to see stuff like this come to fruition.
for another angle - depending on the provider, theyre going to train on these queries and responses, and i dont want folks training an Epstein LLM, or accidentally putting Epstein behaviour into LLMs
It links to the original documents released by the DOJ.
Also, just like LLMs hallucinate and it's up to the person to decide to commit the code into the repo (and they should be held accountable to that), the same applies to people who use this tool to release fake news.
Of course, we try to apply as many "ground-truthing" techniques as possible.
Journalists of all kinds are using Jmail already for their professional work and we are in touch with them when they give us feedback. For example, we've redacted victim's names that we would've not known except for the work of tons of volunteers and journalists—and yes, this was NOT redacted by the DOJ and should have.
But ofc, this is a thorny trade-off between victim protection and censorship.
IMO it’s (unfortunately) the public’s responsibility to learn the lesson that LLM’s shouldn’t be trusted without double checking the source — same position Wikipedia was in 10 years ago. “Don’t use Wikipedia because it has incorrect information” used to be a major concern, but that seems to have faded away now that Wikipedia has found its place and people understand how to use it. I think a similar thing will happen with LLM’s.
That opinion does not take the responsibility away from LLMs to continue working on educating people and reducing hallucinations. I like to think of it as equal responsibility between the LLM provider and user. Like driving a car - the most advanced safety system won’t prevent a bad driver from crashing.
We also are working on crowdsourcing methods, but it's hard because almost everyone involved in the development of this project is a volunteer that either works for a company already or is a startup founder (me)... so is very tricky to find time.
You don't really need a LLM for that. The discourse around the files is filled with allegations/implications of guilt based on spurious factors like number of mentions.
Our development process has been interesting. Although just Riley and I first made Jmail, it's been really gratifying to see companies, journalists, and fellow developers like Diego rise to the occasion to make this entire suite of apps as high quality and extensive as possible.
Great, I've been looking to be able to contact you somewhere, hardly a better place :) Thanks for sharing the project btw, and for keeping jmail alive, been useful to dig into some stuff.
However, there seems to be some "injected" or "fake" emails, that I cannot figure out why they're there in the first place. For example this one: https://jmail.world/thread/55b91b46ef1e4487bee131a8505e14a4?...
For that example, the first problem is that there is no link to the source file from the disclosed files, which is strange, because most other emails have that. Secondly, this almost seems to be a "sponsored" email or something, as it has an ad in the top right corner reading "Sponsored by Drop Site News", but clicking that just takes you to some general page, with no clear information how that's related to the fake/injected email. Also, what does "Verified by X" actually mean, did they verify the authenticity of that specific email?
There seems to be a bunch of people confused by those emails and wondering where they are coming from, because it's missing the source link like the others, so there doesn't seem to be a good way to verify those emails. Could you maybe share a bit about what's going on?
He was a very famous figure in August 2019, and normal people spammed his inbox with emails like the below
https://jmail.world/thread/4accfb5f3ed84656e9762740081a4579?...
These are all real emails! We can do a better job making this clear to the user.
Why do the rest of the emails provide a link to the online file (original one hosted at justice.gov) but the "sponsored"/ad ones do not?
This for example: https://jmail.world/thread/HOUSE_OVERSIGHT_016203?view=inbox
Has a "View Original Document" which takes you to kind-of-the-source (ideally should have been link to justice.gov but better than nothing I suppose), but for the emails with the Dropsite ad, it seems there is no original document?
Many Yahoo emails do show you the original source, and the original is just an EML file. These were files directly exported from Epstein's Yahoo account. Bloomberg used these EMLs to confirm that the Yahoo emails are real (https://www.bloomberg.com/news/newsletters/2025-09-12/epstei...).
We don't tamper with these EMLs, so we currently don't make the EML accessible if the team had to redact any contents of that email.
See this for example.
https://jmail.world/thread/97d4a52d1df3948368770068262d2aab?...
We can fix examples where there are no redactions yet no EML download is available
This Yahoo dataset, which we helped release after launching the first Jmail, also proved Epstein's connection to Iran-Contra (!!). Now immortalized on Wikipedia (https://en.wikipedia.org/wiki/Jeffrey_Epstein#Financial_trou...)
“The Jmail Suite is an interactive archive of Jeffrey Epstein's emails, documents, photos, and more. Data compiled from the House Oversight Committee, Department of Justice, and DDoSecrets releases”
If I understand correctly, the emails you’re looking at are from a leak that is only accessible to journalists — so Drop Site News (as journalists) have access and have published some.
https://ddosecrets.org/article/epstein-emails
https://jmail.world/jamazon
Makes it unsettling when it then shows orders for books you've also read. Hard to swallow having anything in common with such a person.
I would have also assumed he got his clothes from much fancier places than Amazon, considering his wealth. Can't beat two day shipping I guess.
Altogether his Amazon orders are disturbing in places, but also somewhat humanizing, which is not what I expected.
So about that specific thing, how do you avoid this particular problems with LLMs getting basic things wrong and basically being a lottery if the answer is correct or not? It's the reason I've avoided (and encouraged others to avoid it to) using LLMs too heavily in the process of reading, understanding and summarizing documents, but clearly you've must have figured out some trick to this?
I'm also pretty sure these docs are already being used for training, whether or not Jmail / Jemini exists.
Also, just like LLMs hallucinate and it's up to the person to decide to commit the code into the repo (and they should be held accountable to that), the same applies to people who use this tool to release fake news.
Of course, we try to apply as many "ground-truthing" techniques as possible.
Journalists of all kinds are using Jmail already for their professional work and we are in touch with them when they give us feedback. For example, we've redacted victim's names that we would've not known except for the work of tons of volunteers and journalists—and yes, this was NOT redacted by the DOJ and should have.
But ofc, this is a thorny trade-off between victim protection and censorship.
Disclaimer: I actively work on jmailarchive!
IMO it’s (unfortunately) the public’s responsibility to learn the lesson that LLM’s shouldn’t be trusted without double checking the source — same position Wikipedia was in 10 years ago. “Don’t use Wikipedia because it has incorrect information” used to be a major concern, but that seems to have faded away now that Wikipedia has found its place and people understand how to use it. I think a similar thing will happen with LLM’s.
That opinion does not take the responsibility away from LLMs to continue working on educating people and reducing hallucinations. I like to think of it as equal responsibility between the LLM provider and user. Like driving a car - the most advanced safety system won’t prevent a bad driver from crashing.
Also, feel free to check Jwiki (FKA Jikipedia) at https://jmail.world/wiki
As in, OpenAI, Anthropic, and Google's models won't follow instructions regarding forensics for this?