• 0 Posts
  • 8 Comments
Joined 6 months ago
cake
Cake day: May 18th, 2026

help-circle




  • So my setup rigjt now:

    • vps with headscale
    • all devices connect to it
    • in my home lab a vm running a tailscale client and a reverse proxy with a Wildcard cert
    • all other services run as their own VMS and the proxy routes to them
    • i have a dmz (not yet public) running a proxy with tailscale again for a different domain
    • dmz runs a authentik vm which i use for access controll and user management
    • backup Server connects to tailscale as well and proxmox pushes zfs snapshots every day

    My long term goal is to move to pangolin as a reverse proxy on the vps and tailscale alternative and remove tailscale, as i find its very battery inefficient on Android (no subnet Router etc).

    So i just give people access to the VPN and send them an authentik invite and thats it. With pangolin i dont even have to give them a VPN access anymore, it can be done via authentik i think.

    A vps is 3-6€ and the tailscale standard is 10€ so its even cheaper, but a bit more setup to maintain. But it allows a ton of users :) so i guess this would be the smallest change with the highest impact

    Authentik is nice to have but a pain to set up as the oicd stuff can be tricky. Immich was super ez, nextcloud was OK, docspell was a pain and ocis does not work with authentik becsuse they use different architectures for sso. (Ocis and authentik are a bit dumb here)



  • I don’t know how exactly they are trained to behave, I would assume the goal of the creators is to have an AI that is easy to work with so more people use it. That makes them agreeing and write like humans maybe? And for sure they are build to seem smarter and more capable than they are, this is an issue we have to overcome. I suppose its difficult to train an AI to say “I don’t know” because its just a mushy mess of matrices and vectors and data is not “stored” in their system but generated to represent something that seems cmas close as spossible to the truth?


  • Tbh i don’t quite get it.

    I expect to be downvoted like hell here for saying this, but I think discourse is important. Also, its not meant as an attack on op, I want to get my viewpoint out here for others to respond and spark a discission I hope

    You went to chat with one of the lowest their models they offer. It detected an typo and reported it by saying “I made a typo”. You then tried to get it to admit it does not have fingers which it does by saying “what ever the equivalent of a typo is”. So first we don’t habe the language to say an AI generated an false word or token (other than this, which is pretty clunky), second what did you expect from an randomized predicting machine that drew its statistical data from language?

    And why do you use it in the first place if you think its dumb?

    Maybe this community is just meant as a place to rent and let out emotions or preach to the choir, then I am sorry to slide in here. But if we want to change how AI is used and works we have to find real issus and datapoints that we can use to discuss AI usage. There are enough talking points other than the language it uses, like education, energy, environment, noise, capitalism, human nature and a ton more

    I just miss good discussions about political things like this. I would love if we could have a culture on Lemmy again where we upvote good discussions if we agree or disagree. Trolls and dumb arguments or insults can be downvoted. But we will not develop as a society if we just downvote what we don’t like to hear, its not black and white I think. Idk