mrcompletely,
@mrcompletely@heads.social avatar

As a Mastodon admin/moderator I'm not ready to advocate for blocking threads at the instance/server level bc I think a lot of our users will want the connections. But I am going to block the Threads domain personally over AI/LLM data harvesting/training set concerns. I don't want Zuck training his robots off my posts. I also don't trust him at all, and am more skeptical of the integration the more I think about it.

mielcarz,

@mrcompletely Question – since I can see heads.social (and shakedown.social)l on the open web can't it be used for LLM training anyway? It'd be nice if copyright law required opt-in for training but at the moment I don't believe it does? (I could be wrong about this)

mrcompletely, (edited )
@mrcompletely@heads.social avatar

@mielcarz good question I was just discussing yesterday. My understanding/belief is that the difference is one of consent. The big LLM companies are getting regulatory pressure and major class action lawsuits over lack of consent in gathering data for their training sets, which was done by scraping the open web as you describe. Federating with a service that gathers that data probably counts as giving consent, while actively blocking federation should count as negatively affirming (saying no).

mrcompletely,
@mrcompletely@heads.social avatar

@mielcarz so I think for them to scrape data from an account or server that has defederated from threads could create a lot of legal exposure of the kind they're specifically trying to get away from. One reason X and Threads have any value at all rn is their use as training sets in an affirmed consent context (terms of use that specify your use of the service amounts to consent)

Not a lawyer, etc so all of this is provisional interpretation

paninid,
@paninid@mastodon.world avatar

@mrcompletely @mielcarz

There has to be lawyers in the who can speak to this legal exposure.

mrcompletely,
@mrcompletely@heads.social avatar

@paninid @mielcarz yes, I'm waiting for the information space here to clarify, we're in muddy waters rn

Which is fine, just move slowly

mielcarz,

@mrcompletely Thanks for the response. I think the convenience of being able to follow more normies/news sources here may outweigh whatever concerns I have about my content being used for LLM training but it's definitely something to think about and something that other folks may weigh differently.

mrcompletely,
@mrcompletely@heads.social avatar

@mielcarz that's why I think the right thing to do now is avoid blocking at the instance level and inform users that they have the choice to do so individually

arghdos,
@arghdos@heads.social avatar

@mrcompletely @mielcarz I’m extremely skeptical of new regulation / copyright use at this point because these companies have already gotten the benefits of content they don’t own (see e.g., https://aicopyright.substack.com/p/the-books-used-to-train-llms) but new regulations in our octogenarian run country will probably be written by, and for the benefit of these companies (see e.g., https://www.techdirt.com/2023/12/04/if-creators-suing-ai-companies-over-copyright-win-it-will-further-entrench-big-tech/) to literally create the moat Google was moaning about not having.

Maybe the EU will do something halfway decent?

arghdos,
@arghdos@heads.social avatar

@mrcompletely is this the “upload a one line CSV with a single domain to a weird place in the settings” thing I saw and misplaced yesterday?

mrcompletely,
@mrcompletely@heads.social avatar

@arghdos I have no idea what that is

arghdos,
@arghdos@heads.social avatar

@mrcompletely it was the import method @clifff mentioned:

https://medium.com/@geofbard/how-to-block-server-domains-in-mastodon-899b24f8fb6e

Waaaay easier to navigate to the right profile and block a domain from my phone than to convince iOS to let me change and extension

clifff,
@clifff@shakedown.social avatar

@arghdos @mrcompletely Didn’t see that but there probably is a way to do it through csv import. You can also block domains from the website and some apps (screenshots of how to do it in this post) https://shakedown.social/@clifff/111579512518025817

BEK97,
@BEK97@shakedown.social avatar

@mrcompletely I understand very little about the technology behind all this. Will it be possible to set up a one-way connection so I can follow threads accounts (e.g., bands, news) in my mastodon feed without replying or sharing my data with meta? Essentially a version of bird.makeup that works in real time.

mrcompletely,
@mrcompletely@heads.social avatar

@BEK97 good question - it's too early for me to answer though. This is very early days

zakyfarms,
@zakyfarms@heads.social avatar

@mrcompletely @BEK97 that’s what I’m hoping for. Otherwise I’m likely to block.

rowjimmy,
@rowjimmy@shakedown.social avatar

@mrcompletely seems reasonable. Haven’t done that quite yet as I’m taking a moment to think about who/what/why I might want to follow over there.

mrcompletely,
@mrcompletely@heads.social avatar

@rowjimmy yeah I need to really evaluate the value of it which I don't have time for right now

boelder,
@boelder@shakedown.social avatar

@mrcompletely

Same, and for the same reasons. Already done...

mrcompletely,
@mrcompletely@heads.social avatar

@boelder nice, seems like the right midpoint for now. Shoutout to the flexible Mastodon filtering tools yo

  • All
  • Subscribed
  • Moderated
  • Favorites
  • random
  • DreamBathrooms
  • InstantRegret
  • ethstaker
  • magazineikmin
  • GTA5RPClips
  • rosin
  • modclub
  • Youngstown
  • ngwrru68w68
  • slotface
  • osvaldo12
  • kavyap
  • mdbf
  • thenastyranch
  • JUstTest
  • everett
  • cubers
  • cisconetworking
  • normalnudes
  • Durango
  • anitta
  • khanakhh
  • tacticalgear
  • tester
  • provamag3
  • megavids
  • Leos
  • lostlight
  • All magazines