• 19 Posts
  • 756 Comments
Joined 2 years ago
cake
Cake day: March 19th, 2024

help-circle
  • I don’t agree. LLMs are by design probabilistic. Chainsaws aren’t designed to be probabilistic, and any functionality that is probabilistic (aside from philosophical questions about what it is possible to be certain about, YKWIM) is aimed to be minimised. You’re supposed to be able to give the same model the same prompt twice and get two different answers. You’re not meant to be able to use a chainsaw the same way on the same object and have it cut significantly differently. You’re inherently leaving much more to chance by using LLMs to generate code, and creating more work for yourself as you have to review LLM code, which is generally lower quality than human-written code.


  • Not comparable at all. Power tools work deterministically. A powered chainsaw is not going to have a 0.1% chance of chopping a completely different tree on the other side of the forest. Of course accidents happen; your hand can slip. But a proper comparison would be if you got a computer to look at a large number of powered chainsaws and then generate its own in CAD based on what it’s seen, and then you use that generated power tool. Which, for something as potentially dangerous as a powered chainsaw, you most likely wouldn’t want to do, and would want to have careful human oversight over every part of design.






  • The relevance for me personally is whether or not they can be useful for programming, and if they’re accessible to run locally. I’m not interested in feeding my data to a datacentre. My AMD GPU also doesn’t support ROCm so LLMs run slow as fuck for me. So, generally, I avoid them.

    LLMs consistently produce lower quality, less correct, and less secure code than humans. However, they do seem to be getting better. I might be open to using them to generate unit tests if only they would run faster on my PC. I tried deepseek, llama3.1, and codellama; all take like an hour+ to answer a programming question given that they are just using my CPU, as my GPU doesn’t support ROCm. So really not feasible for anything.

    Depending on what you count as AI, I think some of the long-existing predictive ML like autosuggestions based on learning your input patterns are fine and helpful. And maybe if I get a supported GPU I won’t mind using local LLMs for some things. But generally I’m not dying to use them. I can do things myself.





  • communism@lemmy.mltoTechnology@lemmy.worldBitwarden 100% price increase
    link
    fedilink
    English
    arrow-up
    6
    arrow-down
    1
    ·
    12 days ago

    I’ve had my VPS exposed to the internet for a while and never been pwned. No professional experience. Use SSH keys, not password authentication. Use FDE if physical access is in your threat model. Use a firewall to prevent connection on internal-only ports.

    Vaultwarden will store your passwords encrypted (obviously) so even if your database does get stolen, the attacker shouldn’t be able to read your passwords without your master password.


  • It’s great. I also self-host my own Forgejo (that’s the software Codeberg runs on) instance for private repos, to avoid using up space on Codeberg’s servers.

    Main problem is the lack of federation, leading to splintering across Codeberg/GitLab/sourcehut/self-hosted forges. I know there’s Radicle, and Forgejo is working on ActivityPub integration, but it’s slow-moving to get what should be inherently federated by design (git) to actually be federated. In practice you need accounts on a dozen different websites if you want to regularly contribute to foss.






  • Signal is fine for normal/social chatting. It is centralised which makes it much harder to obscure identifying conversation metadata, and I wouldn’t recommend it for comms with a state threat model. I like SimpleX for addressing those issues.

    If you just want to chat to friends and nothing else, I probably would recommend Signal for the most polished experience and most widely adopted open-source private messenger.