moved from piefed.social/u/rozodru due to qutebrowser issues.

  • 1 Post
  • 319 Comments
Joined 4 months ago
cake
Cake day: February 6th, 2026

help-circle
  • Just go with CachyOS. it will allow you to install hyprland during the install process and provide you with Cachy’s dotfiles of it and everything you need right off the bat which you can then further configure yourself or just clone the omarchy dotfiles if you so wished.

    EndeavourOS is fine but I always had issues installing it via ventoy (i.e. it doesn’t.) so I can’t recommend it. With endeavour you’d have to manually install hyprland, manually configure it from the get go, manually install kitty, etc etc etc. Cachy will just save you a bit of time and give you a solid base to start from for everything. Plus the CachyOS kernel is pretty good which is an added bonus. I use it on my NixOS system.







  • remember kids GIGO “Garbage In, Garbage Out”.

    the general public just isn’t aware of this. they assume that anything and everything an LLM spits out is fact. Again, the LLM/AI MUST provide a solution. any solution. that solution doesn’t need to be correct or even exist but in order to retain users it will and will consistently and constantly provide a hallucinated solution.

    And in most cases, without their knowledge, the user is also to blame. I can see a lawyer asking an LLM “are there any cases where such and such happened” trying to find something INCREDIBLY specific to what they need and the LLM 8 times out of 10 will say “yes! there is!” and hallucinate a case that the lawyer needs. You see it all the time if you ask Claude with assistance in trouble shooting a specific tech or code problem. “I’m having issues with this and this, can this library or app or program solve this for me?” “Yes! here’s how” and then will tell you something the library or program or whatever can’t even do but you’ve lead it to that direction and it MUST provide a positive retention-able solution.

    Even using AI as it SHOULD be used is no longer good. just using it as a sort of rubber duck doesn’t help anymore because of GIGO. the training data, across ALL LLMs, is so tainted now you might as well take all the money we’re wasting on data centers and tokens and just toss it at the crazy dude outside the bus station and just ask them the same questions.










  • bingo. I live in Toronto and i’m a contractor that is doing code reviews with a focus on LLMs for startups and other small tech firms. I’d say 9 out of 10 times my reports can be summed up as “this could have all been avoided if a team of devs had remained on staff” and then I fix it.

    They’ll keep trucking along with AI and then hiring people like me for a premium because in the long run it’s still cheaper than having a team of 5+ devs on staff and that line will go up. the AI isn’t improving anything, it’s hindering them but it’s still slightly cheaper than having humans on board. broken and delayed product be damned, their still saving a couple nickels.