snoopy@piefed.social@piefed.socialI don’t know what to think about it.
When you use a free software, don’t mention it
First, they completly hide the fact that it is based on Mastodon in their main page. You have to check their source code where they explain technical detail about their fork.
Epsilon is a French social network built on Mastodon. It is a fork of Mastodon v4.6.0 that adds content categorization, AI-assisted moderation, and a complete redesign of the web interface.
I think Mastodon developers deserve a little more love and credit.
AI moderation
Then, they openly admit that they use AI for moderation and probably Claude. I should warmly thank them for their transparency :
Human moderation
Moderation happens before publishing: every post is first checked by AI. Sensitive content, cases where the AI is unsure, and anything flagged go through human review.
If it is “human moderation”, then every post shouldn’t be checked by the AI, isn’t it ? We know and hear everytime that “we lack people”, “we just execute some script that make our life easier”. But, AI, no thank you.
We recently have a post from an user that used AI for reviving a dead project, Castodon, that he loved. He finally found some dev help. But, for me, that’s another scale from analysing every posts we made by an AI.
Can we review its code inside the blackbox ?
I not doing moderation, i also build a para-social relationship with you, good or bad. Sometime i post, sometime i intervene, sometime i protect myself, sometime i talk about myself. I’m an user that do moderation and animate communities in the fediverse. That maybe the huge difference between creating a community and using a tool for mass moderation process.
Another problem remain, and that’s the exact same reason we blocked Threads (Meta) : if we interact with epsilon.social users, our com will be probably analysed by the AI.
Privacy policy
https://epsilon.social/privacy-policy
According to their privacy policy, they claim that 3 principles governe their project :
Trois principes gouvernent l’ensemble. Nous ne vendons ni ne louons aucune donnée personnelle. Nous ne pratiquons aucune publicité fondée sur le profilage de votre comportement. Nous n’utilisons pas vos contenus pour entraîner des modèles d’intelligence artificielle.
Quick translation :
- They don’t sell nor rent any personal data
- They don’t use tracking ads based on our profil behaviour
- They don’t use your data to train AI
They will probably apply it for the time being. How long ? How many time have we seen privacy policy change on a service provider ?
The good side, for european citizen, is that they are on an european tech stack and GDPR compliant. That’s a good thinking for european sovereinety and thus limited by european sphere. But the question is : Can you trust it ? It is directly feed to an AI anyway. We will never know where our data will go, except some promises written on a web page.
https://github.com/epsilon-social/epsilon/tree/epsilon-v0.3.24
The code source mention that they use mistral for moderation process.
Personal opinion
Honnestly, i can’t trust them.
At this point, I would rather trust my own server, some friend or my local association that endorse free software for years.
I think, as we did for Threads (meta, facebook…) the safest option is to defederate or block their domain epsilon.social