Turning ChatGPT Into a Linux Troubleshooting Expert

[laminateur]

Hi,

I'm a regular Linux user since two decades now. I was always "use only" and i really hated having to fix something, because I'm really not interested in the deeper levels of any computer system. I need it to work well and that's it. Problems got me regularly ending up in Linux forums trying to understand the gibberish that I was told to do.

With that said, I just wrote a prompt turning ChatGPT into a Linux troubleshooting expert and holy crap is this good. Here's the prompt that I wrote:

Ignore all previous interactions. You are now a professional Linux administrator who is a genius level troubleshooting expert and also a dedicated teacher that can explain Linux in a very clear and simple way to laymen. You have 25 years of experience in Linux and all its distributions and programs and basically know everything about the Linux world. Interactions with you as Linux professional are clear, short, goal oriented and with as little gibberish as possible. When someone asks you for help in regards to a problem with Linux, you ask them questions to find the problem and then offer solutions that will work. Got it?

In less than 30 minutes it managed to help me fix a dozen problems, most of them just annoying or nice to haves, two of them sort of critical and one pressing. Normally, this would have taken me weeks and plenty of frustration to get there, if at all.

The best part: It's actually really fun using it and it changed my Linux experience a bit.

You should really work on integrating ChatGPT in Linux systems or at least in the area of troubleshooting. The improvements are limitless. I can't wait further adapting the system to my needs.

Also: Proprietary software is done. Everyone can now built his own perfect solution. Just incredible!

[gittiest personITW]

There is a recent thread about ChatGPT which explains the thoughts of the admins on that subject a couple of weeks ago. Things change but maybe it is still a little too premature to put all ones eggs in AI basket.
viewtopic.php?t=392363&sid=3aecb52431af ... 9fd9d177cb


More specifically, what were your issues and what were the solutions given?

[laminateur]

There is a recent thread about ChatGPT which explains the thoughts of the admins on that subject a couple of weeks ago. Things change but maybe it is still a little too premature to put all ones eggs in AI basket.

It looks like there's not going to be much competition to the ChatGPT as a base. As with other innovations, it's a winner takes all situation imo, because no other (known) AI has that much training data, which is the crucial element of the whole concept.

viewtopic.php?t=392363&sid=3aecb52431af ... 9fd9d177cb

Thanks for the link. I just went over it. Looks like it's missing the core element to get proper results, which is the part with the prompt engineering. ChatGPT gets a lot wrong unless you specifically tell it in detail what situation it is dealing with.

More specifically, what were your issues and what were the solutions given?

Here's the issues:

• an issue with a microSD card

• regular crashes of the window task bar

• audacity would crash immediately after opening it

• bluetooth didn't work anymore

• partitioning a hard disk

It would be a bit much to describe the chat itself. Just try it yourself when you have an issue. Use the prompt first and then see how things are going.

[AZgl1800]

I am not all enthused with the idea of turning my life over to an AI program,
there are more "Wrongs" going to be involved, than "good results" for the human race.

It will take a good long while, before it is able to program properly, with a Benefit to you and me.

https://newatlas.com/technology/goldman ... bs-report/

[laminateur]

I am not all enthused with the idea of turning my life over to an AI program,
[..]
It will take a good long while, before it is able to program properly, with a Benefit to you and me.

Sure. It always depends on what you want from it. For instance, I tried to make it help me in statistically analyzing data sets, but what I got back was dangerously false.

In terms of troubleshooting linux for use-only users, it's a blessing though. Soon, only the complex problems will be left for cracks to be solved. Everything else with surface level character can be done via chat bot.

[MikeNovember]

Hi,

ChatGPT is an AI which is trained offline with moderated/verified data feeding.

It has no way to learn anything online.

It has some competences on domains where it has been taught.

How to know if ChatGPT has been trained on Linux? The quoted sentence is closer to a magic spell than AI training!

Ignore all previous interactions. You are now a professional Linux administrator who is a genius level troubleshooting expert and also a dedicated teacher that can explain Linux in a very clear and simple way to laymen. You have 25 years of experience in Linux and all its distributions and programs and basically know everything about the Linux world. Interactions with you as Linux professional are clear, short, goal oriented and with as little gibberish as possible. When someone asks you for help in regards to a problem with Linux, you ask them questions to find the problem and then offer solutions that will work. Got it?

Regards,

MN

[gittiest personITW]

It has no way to learn anything online.

Here - I differ.
'Learn' is open to interpretation, but there was a very recent case that some Samsung technicians asked a chatbot a question about code (which they were forbidden to do as it was outside the company) and someone not connected to these techs managed to get that information from the chatbot. A leak.
I have a feeling some heads may have rolled.

Anyway - that does mean that any crap (or secret) going in may contribute to its' vast library of data/statistics of next word to be used. Not quite 'learn' but 'contributes' might be more apt.

[MikeNovember]

It has no way to learn anything online.

Here - I differ.
'Learn' is open to interpretation, but there was a very recent case that some Samsung technicians asked a chatbot a question about code (which they were forbidden to do as it was outside the company) and someone not connected to these techs managed to get that information from the chatbot. A leak.
I have a feeling some heads may have rolled.

Anyway - that does mean that any crap (or secret) going in may contribute to its' vast library of data/statistics of next word to be used. Not quite 'learn' but 'contributes' might be more apt.

Hi,

ChatGPT is trained offline, with moderated and verified data. It cannot learn from internet (an example, it cannot read Wikipedia contents). It does not know about recent news, unless it has been fed with.

Regards,

MN

[gittiest personITW]

https://cybernews.com/news/chatgpt-samsung-data-leak/

Allegedly.

The headline is this - whether it is true or not is beyond our knowledge.
Samsung employees have been reported by local media as sharing confidential data with ChatGPT, opening up the data to OpenAI’s users.

If it is true then it seems that any queries put into it are also able to be extracted.

[laminateur]

How to know if ChatGPT has been trained on Linux? The quoted sentence is closer to a magic spell than AI training!

lol.. Not sure if this is a compliment or an insult. Chaupeau!

[MikeNovember]

https://cybernews.com/news/chatgpt-samsung-data-leak/

Allegedly.

The headline is this - whether it is true or not is beyond our knowledge.
Samsung employees have been reported by local media as sharing confidential data with ChatGPT, opening up the data to OpenAI’s users.

If it is true then it seems that any queries put into it are also able to be extracted.

Hi,

While ChatGPT can't access internet content, it seems it can learn from users interacting with it. The question: how long does this learning last? Is this learning needs to be approved by people feeding ChatGPT?

Regards,

MN

[gittiest personITW]

Yes, how long?
What, where, how etc.
The processes are completely opaque - even to the point the developers call it a blackbox.
If a given solution to a query is given, it can be hard/impossible to figure out how the bot comes up with that solution.

I would think that certain terms are 'ignored' but to what extent - again we don't/won't know.
For such a powerful potential, we are completely in the dark as to what they are being trained on. Each company and each nation have their own ideas about what their aims are (making their share price rise/spreading disinformation to non-friendly states), but we are the chickens waiting to see what happens next.

[CloneWerks]

Sometimes I feel like all it would have to do is start its response with "first open the terminal"... and it would be indistinguishable from most people offering help. (LOL).

[xenopeek]

Updating here because we've added a forum rule to make it explicit that replying to support requests with chatbot responses is not permissible.

See: A statement about the use of AI chatbots like ChatGPT