21 ms·

> just giving the LLM the ssh keys and having at it. It’s pretty good. I'm mostly fine letting AI write code because its easily auditable and you can follow ch

by wiether 11d ago

> just giving the LLM the ssh keys and having at it. It’s pretty good. I'm mostly fine letting AI write code because its easily auditable and you can follow changes through the Git history. Whereas, it you let the AI go free on a server, you mostly have no idea what it did and where, so it's quite hard to audit and even harder to try and fix things. So, unless relying on a declarative paradigm like NixOS does, I'm not ready to give AI the keys to my castles

yoz-y 11d ago
Admittedly this was not my castle. More of a $1.5/mo shed that I got just to try this.
- BetterThanSober 11d ago
  That's a cheap shed, what's the provider and spec?
  - yoz-y 10d ago
    I lied actually. (Misremembered) it’s actually about $2.15 https://www.netcup.com/en/server/vps/vps-pico-g11s-iv-12m-nue https://www.netcup.com/en/server/vps/vps-pico-g11s-iv-12m-nu...
walrus01 11d ago
While admittedly this can be a bit risky, there's a lot less possible blast radius or thing that can go catastrophically wrong if you give the LLM the ssh keys and sudo access on a KVM VM (or equivalent) that's been created as a blank slate exclusively as its playground. I have a template debian 13 KVM VM that can have its ssh host keys and hostname wiped, other unique per-node things, and duplicated with a basic bash script in a couple of minutes to deploy for home lab testing. I'm fine with giving a "smart" self hosted LLM like qwen 3.8-flash-next Q8 access to it to do things in opencode. Don't give it the keys to any castle but give it access to one specific containerized thing...
- wiether 11d ago
  > for home lab testing So you're talking about a sandbox, not a prod server hosting personal data?
  - walrus01 11d ago
    I'm saying a staging/testing environment is a good place to prove your specific configuration of "smart" LLM isn't going to do something wildly wrong, before you even think to trust it in a production environment. And as a place to develop things like playbooks and similar documentation to feed into an LLM's context for ongoing maintenance and operations of something in production.

by wiether 11d ago

yoz-y 11d ago

Admittedly this was not my castle. More of a $1.5/mo shed that I got just to try this.

BetterThanSober 11d ago
That's a cheap shed, what's the provider and spec?
- yoz-y 10d ago
  I lied actually. (Misremembered) it’s actually about $2.15 https://www.netcup.com/en/server/vps/vps-pico-g11s-iv-12m-nue https://www.netcup.com/en/server/vps/vps-pico-g11s-iv-12m-nu...

walrus01 11d ago

While admittedly this can be a bit risky, there's a lot less possible blast radius or thing that can go catastrophically wrong if you give the LLM the ssh keys and sudo access on a KVM VM (or equivalent) that's been created as a blank slate exclusively as its playground. I have a template debian 13 KVM VM that can have its ssh host keys and hostname wiped, other unique per-node things, and duplicated with a basic bash script in a couple of minutes to deploy for home lab testing. I'm fine with giving a "smart" self hosted LLM like qwen 3.8-flash-next Q8 access to it to do things in opencode. Don't give it the keys to any castle but give it access to one specific containerized thing...

wiether 11d ago
> for home lab testing So you're talking about a sandbox, not a prod server hosting personal data?
- walrus01 11d ago
  I'm saying a staging/testing environment is a good place to prove your specific configuration of "smart" LLM isn't going to do something wildly wrong, before you even think to trust it in a production environment. And as a place to develop things like playbooks and similar documentation to feed into an LLM's context for ongoing maintenance and operations of something in production.