It still works, you just have to provide a unique user-agent. There's also x0.at which has a web upload form as well.

It doesn't, pretty sure the repo is a joke. Their code doesn't do anything special, just runs the program you give it.

Yep, the project must be a joke because the code doesn't actually do anything besides run the specified program normally.

Only if your OPSEC was already bad... LLMs aren't magical, they're just piecing together information you already gave away freely.

Also the authors of this paper refuse to show their work "for safety."

I wouldn't blanket call the removal of PFS a "failure" as they put it... it does make the protocol much simpler (and hence easier to understand/audit as well) and it's not always a necessity for every single person's threat model... which is an important phrase the article doesn't even mention.

IMO arguing about security or privacy without both people first defining their threat models... is like claiming apples are objectively better than bananas in every way.

If they did, it's still entirely possible to use your own e2ee for messages between friends on any app just by using an encrypted keyboard app on your phone (like KryptEY).

URLs can have newlines too

I think your own statements here are pretty elitist.

"As a rule, strong feelings about issues do not emerge from deep understanding."

Maybe a better question might be why all the upvotes. Perhaps that already answers one of your questions.

it doesn't

I think the article fails to take several critical factors into consideration.

• The complexity of dealing with such large amounts of information will keep increasing forever as the amount of information also grows

• AI struggles with conflicting information and mistakes, which happen a lot especially when humans are involved, so eventually you will have lots of "garbage in garbage out" issues causing problems

• The data one might be able to track will continuously be challenged or removed on legal/compliance bases over time, reducing its availability

For example: Yes the NSA might want our chatbot logs, but after enough people realize they might be/are getting them, people will stop feeding it as much, or introduce noise on purpose. It's not a perfect vacuum of constant reliable information forever. We are already seeing that AI models learning from web results are getting caught up in their own slop making themselves dumber. And the sheer volume of information relative to the computing power necessary to process everything will also become a problem if they keep trying to process every single thing.