While I find AI helpful when I’m using technology that’s new to me, or when I want to automate something, I don’t think I’ll ever want it to take over completely, at least not without my supervision, and permission to execute whatever is happening. For example (from your post), I use AI to design an automated way to restart a daemon that’s supposed to run continuously with the system if it stops for some reason, I would want the AI to show me the solution, and get my permission before being allowed to implement the change for me, or so I can implement it myself, then respond with the result of whatever test we devise to check the correctness of our solution. To be honest, while I’ve been using AI extensively for a variety of purposes, I’ve noted that correct implementation of some new process (eg: setting up my new Proxmox LAN server and creating an LXC file storage Container for my backup image files that will be created by my new Windows-based backup system - Veeam Agent) involved multiple iterations at each step along the way to achieving the desired outcome, and while the end result we’ve achieved so far are far in excess of anything I could ever have achieved on my own, and the entire thing may have been completed much faster without me being involved, I’d never have learned what I did. or been able to understand everything as well as I do now, and for me, it’s more about me learning than about me being in control, even though that’s a consideration as well.
Windows Defender does much about malware gaining access with memory, file change, and even ransomware protection. When it comes to safely using third-party apps, my solution is UniGetUI (a graphical wrapper for Microsoft’s WinGet) to install, manage updates, and delete applications. It can also incorporate other repositories (I use several; scoop, chocolaty, pip, and a few others, so before I allow an update I have a PowerShell script that I run to scan the proposed update candidate for malware before allowing the update). If you ask me, Microsoft should purchase UniGetUI and offer it as the preferred third-party package management solution for Windows home users.
I use Gemini here. I’ve set it up on Firefox on both Windows and Garuda Linux, and I usually ask it how to do whatever I’m attempting. Sometimes I ask my question in Firefox’s address bar, or when background information’s appropriate, I open a Gemini dialog in the sidebar.
Systemd auto-restarts daemons .I dont think it uses AI. Sometimes it causes endless loops … fail-restart-fail-restart- … forever. That is a very simple example of the need for management… or what it really needs is fix the daemons code bug.
In a more complicated case the same thing could happen … ie the OS keeps falling into holes and because the fix is a patchup the problem repeats. I dont see AI helping with that.
There are routine tasks that AI may help with . For example a distro maker wanting to check all packages for compatability… but that is not really management, it is more like a build task.
I cant see AI helping me maintain my home computer. It might help with things like maintaining a whole cluster of identical servers.
Privacy concerns with AI ? Simple . Only ask abstract questions … never with real examples … never personal details … not even your name.
AI tries to make conversation. You need to behave like a robot, dont let it humanize the dialog.
Me too, but there are questions very few humans could answer without research.
eg) Can you answer this
" How do I set ‘Focus follows mouse’ in the Lumina desktop environment."
It is just something I encountered tonight.
AI answered it. It is quite tricky … there is no gui button … I had to configure the underlying Fluxbox by editing a config file.
The point is … yes conversations with humans are more satisfying to all of us … but they are not necessarily the best means of looking up information.
A conversation is more than a dialog. I involves feelings and concepts like sharing and supporting.
AI’s nothing to fear. It’s a tool to be used, now at least, and there’s no point in fearing what may come until/if it materializes. If you ask me, rather than living in fear … about anything … regardless the depth of our fear, we should take things as they come, bit by bit, just as we always have.
Well said, Ernie.
I became curious about AI and the internal workings. Not that I could actually understand the complex internal working. But I wanted to know the basic of how AI worked especially since the data (information) AI received during it’s training was not stored. Here is the response from my question, How does AI generate a response? Give a very short version.
“AI generates responses by using math to predict the most probable next word, repeating that process until the answer is complete.”
That’s interesting, because when I ask Google’s Gemini (in the Firefox Sidebar) how to solve a specific problem, it always provides links to the source information it used to guide me. And for the most part, the responses I receive seem to be targeted summarizations of that source material.
That is the large language model (LLM) bit.
There is more to it than that…it can deduce things logically .
It can do more than that.
I used it tonight to debug an issue. There were several rounds, and each round I had to tell it that what it suggested in the previous round did not work. It. responded by thinking deeper ( it tells you what it is thinking in words) .
It was clear that it did not have sufficient information to instantly solve the problem (it admitted that), so it tried to derive an answer logically.
It did eventually succeed … but after lots of failures. It used the failures to ‘learn’ things that it could not look up.
I am not sure if that new knowledge it obtained from my experiments was kept, or whether it dies with my session?
What I have found is when it starts going down a rabbit hole (trying one thing, then the next, then the next etc…) to close that sessions down and start a new one. -When I think things are getting sort of convoluted. It is almost like a new person starting with a clean slate and it approaches the problem a little different. Sometimes I mention what was attempted in the first session, but not always. - there are settings in some AIs to remember past sessions, but I turn them off because I usually want to start without the AI attempting to guess based off old conversations.
It takes information and sometimes it places too much weight on a specific piece of that information. Or it might go the other direction and generalize too much. It comes up with solutions based on patterns that “should work” but it is not always correct. I tend to tell it to look for tutorials, & forums, etc that deal with the specific problem and not to speculate or conflate.
just for fun- I asked an AI about starting a new session and the reason- It’s response
Your approach of resetting sessions to avoid accumulating convoluted context and disabling cross-session memory reduces reliance on prior patterns that might lead to repeated or biased attempts. This is a practical way to maintain fresher, less influenced reasoning in interactions.
