Important: Please do not run commands generated by AI on your servers

mth

Next year in the roadmap: AdamGPT 😃

DracoBlue

Maybe Adam will create a prompt for ChatGPT? hahaha 😃

CloudyBlake

Am I the only one who saw this and thought "yeah, and don't play in traffic also"

Like, I'd never do that because I barely trust AI to tell me the time. But I'm also an old curmudgeon.

xyzulu

CloudyBlake You put it a lot nicer than I would have 😭

ordinary

I'm gonna use AI to build my own panel, with blackjack and hookers. Who's with me?

JohnB

xyzulu Yeah its the same level as an idiot using chatGPT to make a diet that ends up nearly killing them.

Its essentially a sales rep that's been told to say whatever the customers want to hear in a way that sounds authoritive and responsible. And then when you slip on the wet floor and break your leg, the sales rep goes "omg you're absolutely right to be upset. No worries, lets try a new approach and jog upstairs where I can help you better" ... "Ahhh you are wise to point out ur broken leg. Instead of jogging lets try running up the stairs so that I can better help you"

Maybe I'm just arrogant but I feel like you have to be an idiot to trust chatgpt the same way you're an idiot to trust a salesperson who lied to you or talks over you.

Tino

My bad

Tino

JohnB I use ChatGPT to help point me in right direction 😎

cPFence

Tino

It’s more like a sketchy map than a chauffeur, helpful sometimes, but don’t ever hand it the keys.

standtech

This is also a THING now. People using random commands on their servers.

rdbf

standtech This is also a THING now. People using random commands on their servers.

Which they always did, but at least the commands came from people who mostly knew what they were doing, like on Stackoverflow.

Then SEO made it nearly impossible to find what you were looking for, with users ending up on slop tutorial sites written by people aggregating useful code snippets with minimal context for financial gain by using expert level SEO. Then LLM's came along, swallowing it all, spitting out this plausible sounding stuff with zero accountability or possibility to check if any of it is true, possible, dangerous, harmful or whatsoever, while being very accessible to anyone who's looking for answers.

I don't blame anyone with less than system admin level of knowledge for trying stuff that AI spits out when dealing with issues..... Which is probably where stellar extensive documentation, which gets absorbed by AI as well, can make a big difference in the amount of crap that support has to deal with.

JohnB

rdbf Which they always did, but at least the commands came from people who mostly knew what they were doing, like on Stackoverflow.

Indeed. Witnessed first hand someone in a stackoverflow chat really abused the "free" work and would just copy and paste whatever you gave them. They were told this is a very bad idea but didn't listen. They'd also skip steps in instructions and waste your time.

Eventually someone made a legit looking script that they had been begging for but what it actually did was install some "malware." Any time they logged into SSH it would give a popup like "you shouldn't run scripts you don't understand. goodnight" and then reboot lol.

Guess who came to chat asking how to fix it...

"You made the backup I told you to make first right?"

He had not...

slimx

I just put this in my PROJECT INSTRUCTIONS

`Enhance — do NOT use Docker

No Docker. Enhance does not use Docker for platform services or websites. Do not suggest Docker commands, images, Compose files, or Docker networking for any Enhance task.

OS packages + systemd. Services are installed via the OS package manager and managed with systemd. Use native Linux tooling for start/stop, logs, and updates.

Per-site containers (not Docker). Each website runs inside an Enhance-managed site container using Linux namespaces/cgroups and overlayfs. Isolation is at the website level.

Reverse-proxy model. The host web server reverse-proxies into each site container. App ports bind inside the site container, not globally.

Roles + agents. The control panel orchestrates; role servers run an agent that builds/configures runtimes and websites.

Backups/cloning/staging. Operate at the whole-site (container) level, not per addon domain.

Troubleshooting. Use systemctl, journalctl, package manager commands, and Enhance paths under its container directories. Never reference Docker daemons, images, or iptables Docker chains.

If a suggestion involves Docker for Enhance: it’s wrong.`

Works well