The aim is to learn. By being very selective about training we should be able to limit its size and sphere of knowledge. Then we may be able to see how it deals with questions, because we would have inside knowledge of what it knows.
Have you seen any software that would allow us to train up our our LLM? It would mean statistical fitting of parameters to data.
I use CoPilot and the Google AI when building LFS, some things I agree with and some I do not, but keep the questions simple and on target, for the best results!! One has to also rely on one’s own knowledge, to get the needed results!!
Well said.
Could a poorly educated person use AI to build LFS … No.
With enough patience and determination one will eventually build LFS!! AI only gives answer’s the user has to apply the proper commands!!
That requires some Unix or Linux education.
I have no idea. All I can do is describe my experience using the AI Agents I have experience with. Sorry!
Ernie
When I don’t really have direct experience with whatever I’m trying to accomplish, I explain what I’m trying to accomplish with a prompt that’s as detailed and concise as possible, along with requesting a step-by-step guide, then I execute one step at a time, posting the result, regardless the outcome. That way I keep Gemini as on task as I can, and when we don’t get the expected outcome and I post the result, Gemini returns an adjusted command and further steps that will be compatible with the changes. I’ve achieved excellent results using this procedure, even if it takes longer. I, for one, am not concerned with the amount of time required to achieve a positive result. For me, it’s the result that counts.
Ernie
I disagree. With great desire and perhaps the help of a GNU/Linux user who has completed the task, I say it could be possible, albeit unlikely, and the scenario presented raises the question ‘Why would such a person prefer to go through all the effort of creating an LFS system would require when there are plenty of excellent distributions available.’
Ernie
Or the ability to request a step-by-step guide from your favorite AI Agent, and following it one step at a time. That’s how I got my LXC Container created that I use to store my Veeam Agent backup images on my Proxmox home lab.
Ernie
I see two paths. One much more difficult then the other.
Path one.
Build your own AI and provide the training. This option will entail a huge amount of effort, yet be the most rewarding in learning AI from A to Z.
Path Two.
Install a tiny / small AI that has already been thru the learning. Being small, it has a limited amount of training and still can be tested.
From my talk with AI about Path One..
And Path Two from AI.
The base LFS build already has a guide, no AI needed!! The why I build LFS, is simple, because I wish to build LFS!! Yes, there are other distros, but none build with open source, as does LFS!!
We probably should try path2 first.
That experience would help with path1
Lets find a small prebuilt model
[quote="Ernest N. Wilcox Jr., post:49,
topic:16043, username:ernie"]
Or the ability to request a step-by-step guide from your favorite AI Agent, and following it one step at a time. That’s how I got my LXC Container
[/quote]
Ernie , you have some considerable amount of Linux education.
When I built Gentoo, I had help from @4dandl4 . I had a previous try that failed. I dont think I would have succeeded without help … there is just too many things to learn all at once.
We did not have AI then. I am not sure if AI would have helped with Gentoo , because there are excellent manuals for getting information. The difficulty comes from having to learn many things all at once … it is a huge hurdle.
When you have done such a thing once, the information overburden disappears. Even if you forget details over time, relearning is easy. If you do it frequently, you dont forget… it becomes part of you. Then you dont need either help or information.