If you feed somebody's posts into your shitty LLM without an explicit and reversible opt-in, you are violating the consent of fedi users.

This "Zeitgeist" app not only violates consent, but it also doesn't have a centralized ActivityPub identity that can be blocked or defederated.

#FediAdmin #Fediblock

https://alpaca.gold/@seldo/116229243453108157

The creator is making pretty clear that it's designed this way to make it harder to block.

Normally, you'd expect an LLM scraper to have a single (blockable) ActivityPub account doing the scraping.

This is designed so that everybody who uses the service has their identities harvested to steal people's data on the app's behalf.

#FediBlock #FediAdmin

Taking the block-evasion one step further, according to the thread the service doesn't set a user agent so it's not possible to block that way.

Again, according to the thread it does specify an ActivityPub client named "Zeigeist".

Would appreciate any feedback from admins about how to block a specific ActivityPub client.

#FediAdmin #Fediblock

The service also does not currently respect any of the various opt-outs to indexing that people use.

Which means it ignores some pretty universal web standards about consent for indexing.

Even if your excuse is that you're a little baby who doesn't know anything, you should know that.

So far it seems like there's no way (short of a lawsuit) to block this completely.

Instance admins can (and should) do web server hacks to block their own users from logging into the app.

@fedilore we just reject requests with redirect_uri back to zeitgeist. This prevents them from registering their app

https://mastodon.pnpde.social/@spielleitung/116296598330620006

spielleitung (@[email protected])

Attached: 1 image Wir blockieren die App von zeitgeist.blue. Das bedeutet, dass kein Account auf pnpde.social die App nutzen kann. Das bedeutet leider nicht, dass eure Postings nicht über andere Server von der App gelesen werden können. Absichtlich kein Link. Es geht um eine Software, die die letzten 24 Stunden der Timeline eines Accounts in eine LLM liest, um daraus eine Zusammenfassung zu generieren. Technisch simpel: Wir rejecten Requests mit entsprechendem redirect_uri-Parameter in nginx: #MastoAdmin

pnpde.social
@fedilore
I'm super thrilled with his, "now I know why I left," attitude about being called on his BS. Seems weird to build a tool for a service he doesn't intend to use... that is unless the main purpose is data harvesting.
@TheGreatLlama That's because the main purpose is data harvesting. 🤷
@fedilore
It's pretty hard to draw another conclusion.
@fedilore Is there any reason I can't just moderation block alpaca.gold entirely? Seems like a single user instance.
@celestiallavendar You should block alpaca.gold, but the app itself is a client which means it uses the user credentials of everybody who uses it to feed the LLM.
@fedilore That makes sense. If it's a mastodon client I wonder if *key instances are safe. Probably not I would assume, cuz the posts still federate.
@celestiallavendar Nope. Anything even slightly compatible with a Mastodon phone app can be logged in like this.
@fedilore Yeah, that's what I figured. Honestly, it makes me think that there should be a way to implement some kind of bot & scrape layer to ActivityPub that poisons all outgoing content for bots while leaving it fine for users. Could solve a lot of these Fedi scrape / LLM / data harvest issues I think.

@fedilore Alt text: screenshot of Mastodon conversation:

Jan Lehnardt @[email protected] Mar 15
@seldo how can I make sure my posts are never used by this app to be sent to an LLM vendor?
.
Laurie Voss @seldo @janl You would need to block me from following you, I suppose.

Jan Lehnardt @[email protected] @seldo you personally or some app account?
(Note that this kind of stuff is not well liked here on mastodon, I got browbeaten into shutting down just a search engine that worked that way, let alone exfiltrate a timeline into an LLM provider)

Laurie Voss @seldo
@jan Me personally, and also anyone who uses the app. It's just a mastodon client, fetching the timeline. It looks like every other mastodon client.

@aral Appreciated. I've updated the post with that alt text.
@fedilore @zzt i hope we #fediblock this instance with prejudice
@fedilore oh wow, this guy is unbelievable 🤪
@fedilore this violates the ethics of our profession