Oh Yeah,
AI was supposed to be in the new moved topic.
I am getting lots of moving practice.
I will ask Sourav.
I think yes, because AI seems to annoy some of us
1 Like
Relax, you have your fiesta today.
Here is a site that might be of interest to those in this topic.
It’s a small AI you can install as a tool that will help teach you what it is doing. It is aimed at Python coders, but may be helpful just for AI education.
2 Likes
Thank you. It looks like that will do what I want
I am happy to use Python. I had some practice with Python when I was doing that image analysis project.
I might use Artix for this , because that is where I have Python setup. …nothing like having a separate distro for each job … its like having multiple tabs in your browser.
A classroom example is exactly what I need.
I love the " \pi is wrong" rant.
3 Likes
I gave this topic an AI tag.
2 Likes
It helps you think about different projects a bit better too. A clean separation can help me think clearly or focus better.
2 Likes
Down the Rabbit Hole
I wanted to learn more about AI so I asked Copilot the following question.
“tiny / small model to test and learn with include subject model trained on”
This started a 10 hour plus of loading, testing, modifying, correcting, looping,
and more back and forth with the AI. At times it it was rewarding to see some
progress and very frustrating to see failure after failure.
Briefly, I wanted an AI that was small that I could run on my laptop and run offline for testing to see what an AI could do. I want to load a small book and see how the AI would respond to questions and searches on this small book. Information I knew but new to the AI.
After many failures and correction / modification Copilot came back to me and said it
could not be done. WTF! Copilot sent me chasing this rabbit and after all those hours said it could not be done.
Copilot said;
“I gave you the wrong picture of what Ollama (the AI on the laptop) can do.”
Copilot sent me down a rabbit hole.
3 Likes
Yes it misled you. You cant do it with llama… not totally at home.
Relax, you learnt a bit along the way.
I am finding the jargon in this area is prohibitive.
Have a look at @pdecker 's Tau. … see reply 63.
You have to use Python.
1 Like
You might want to check on this article.
It describes a few local AI tools that allow you to interact with your documents.
3 Likes
Thanks,
What does “chat with your pdf doc” mean?
Does it mean you use your pdf doc to train an AI model?
Is it generative AI?
Can you question your pdf doc?
We need to test run some of these and learn what they actually do.
There is no substitute for hands on.
1 Like
I think they refer to it more as RAG, Retrieval Augmented Generation. It can retrieve answers to your queries from the documents you provide.
3 Likes
Linus Torvalds apparently used AI.
From an article on ZDNet.
If you want to read more.
2 Likes
I read it. This line makes it clear
" Torvalds thinks AI can be quite useful for programming and maintenance"
So it is being used to write code, and to find bugs.
Apparently AI for bug reporting has improved dramatically over the past year and is now seen as a valuable tool.
The other interesting comment in that article is that rust usage seems to be settling in.
My gut reaction was against all of this. I guess I have to adapt. It does not change my rejection of systemd … that is a separate smaller issue.
2 Likes
Something that might be of interest to some and related to @pdecker posting of ‘5 Local AI Tools’. This option does not run local, but you can load a document into an AI and have it only search on the file you upload. As a test, I uploaded some fables to Copilot. The txt file can be save in one of your saved chats so the txt file would not have to be uploaded again. Here was some of the conversation. The ‘-’ lines are my input. Edited to make the conversation more readable.
Ready.
-how many fables
93
Ready for the next thing.
-name of 1sr fable
The Cock and the Pearl
-name of last fable
The Fox and the Goat
-largest fable
The Town Mouse and the Country Mouse is the largest — longest text in your attached book.
-shortest
The Mountains in Labour is the shortest.
-okay to ck web on this q. which of the 93 fables is most popular
Most popular of the 93:
The Tortoise and the Hare.
Tortoise and Hare wins.
-what does the story try to tell some one
For the most popular fable — The Tortoise and the Hare — the message is simple:
Slow, steady effort beats careless speed.
If you want another fable’s message, say the title.
-summarize tortoise n hare
Tortoise and Hare — summary:
The Hare races the Tortoise, sprints ahead, gets cocky, takes a nap.
The Tortoise keeps moving and wins.
If you want the moral again, say next.
-next
Moral:
Slow, steady effort wins; arrogance loses.
Ready.
-summary fable 10
Fable 10 summary (from your 93‑fable text):
The Wolf and the Lamb — A Wolf invents excuses to blame a Lamb for imagined offenses, then kills him anyway.
If you want the moral, say next.
-moral
Moral:
The strong will always find a pretext to oppress the weak.
-next fable
Next fable in your 93‑fable sequence after #10:
#11 — The Bat, the Birds, and the Beasts.
If you want summary, say next.
-what word is used most often
Across your 93‑fable text only, the most frequent word is:
“the.”
If you want the most‑used noun, say next.
-what word is used the least
In your 93‑fable text only, the least‑used words are one‑off proper names that appear in a single fable and nowhere else.
Examples:
Zeus
Jupiter
Phoebus
Each occurs exactly once.
If you want the least‑used common word, say next.
-next
Least‑used common word (from your 93‑fable text only):
A single‑appearance everyday word: “yonder.”
If you want the next rare common word, say next.
-I remember a fable about a fox and a bird can you find it
In your 93‑fable text only, the one with a fox and a bird is:
“The Fox and the Crow.”
If you want summary, say next.
-summary
Summary — The Fox and the Crow:
A Crow holds cheese.
A Fox flatters her, praising her “voice.”
The Crow, proud, opens her beak to sing.
The cheese drops.
The Fox takes it.
If you want the moral, say next.
-moral
Moral — The Fox and the Crow:
Don’t fall for flattery; it’s usually a trick to take something from you.
-what is max .txt file that can be loaded
There isn’t a fixed “.txt size” number I can give you here.
Practically, for good results:
Pasted text: keep it to roughly a few hundred pages (low‑MB range) or less.
-what other kind of written input can be updated for reviewing
If you mean “what other kinds of written stuff can you load for me to review the same way as your fable .txt?” — then anything that is plain text works:
notes
journals
letters
stories
logs
scripts
documentation
Basically: if it’s text, I can ingest it and work with it.
If you want me to check a specific file type, say next.
-pdf?
Yes — PDF works.
3 Likes
My Garuda Mokka Linux distribution uses systemd. A while back, after missing a few doctor’s appointments because I was so engrossed in what I was doing, that when I looked up, I realized the time and that I missed yet another appointment, so I decided that I needed something to alert me when it’s time to stop what I’m doing, and take care of whatever needed doing. None of the reminder type apps I’ve found, simply put up something like a toast note for a few seconds, then they’re gone from view. I need something a bit more … ‘in my face’, so I used AI to vibe-code a Python script to sound a short beep, and put a small window on my desktop that gets its title and a message to display in it’s body from the command line, and that works on both Garuda Mokka Linux and Windows 11 Pro 25H2. It took quite a few iterations to get a script that runs reliably on both Garuda and Windows, but we (Gemini and I) got the job done, and my script works very well. On Garuda, I have a SERVICE template file for my script, accompanied by a TIMER file for each reminder I schedule. These are all stored and executed with my user account’s privileges/permissions, and the files live under ~/.config/systemd/local. Once I got my first reminder set up and working, the rest were a simple matter of copy the content of an existing TIMER file, paste it into Kate, edit the title, message, and time parts, then save the new file with an appropriate and unique name. With that done, I refresh systemd, enable the new job, and refresh systemd again so my new reminder runs on schedule. The best part is that if the scheduled time for any repeating reminder passes while I’m in Windows, or the computer’s shut down, it runs as soon after the scheduled time passes as possible. I wouldn’t be able to do that with cron, and I don’t know of any other process that would do what I need as well, so no one has to like or use systemd if they don’t want to, but it has been very useful for me!
1 Like
Interesting.
That sort of negates the idea of doing it locally.
but
It interprets your text presumably using its LLM training … it knows more than just the text … eg it is able to infer morals.
So you did not train it with the text … it is using the text as a sort of giant question.
1 Like
Have a look at anacron.
You dont need an init system to set a timer.
1 Like
Anacron operates based on days, but my reminders run at scheduled times of day, some are one-time job, others run daily at a scheduled time, and a few run twice daily, e.g.: at scheduled times in the morning and evening, but not all run every 12 hours, one runs at 8;00 am and 10;00 pm
2 Likes
That is correct. The LLM was not trained on my text file.
That is one of the most interesting things I have learned from what I have read about an AI. Training of the LLM only happens once when the AI was put together with its many layers.
The LLM already knows how to read, interpret, and answer. A text file is only input to this process that the AI already knows. When I upload a file, the retrieval system in the AI finds the information that matches my question, and the LLM reads that chunk of information using the abilities it learned during training.
An example I was shown went like this: A person does not have to be trained to read again when handed a page to read.
2 Likes
Some AI’s seem to retrain regularly.
I was recently asking some questions about antiX, and it fed me back verbatim information from one of my own recent itsFOSS topics.
That is OK… they have to stay up to date.
1 Like