It always seems impossible until its done.
â Nelson Mandela
2 Likes
You mentioned 2 failures:
a fault in a software component, or
a failure to coordinate
What about hardware failures?
I managed large clusters for years. I managed one site that had a small cluster that had been âmanagedâ by the users for about two years, all engineers, all smart and some with desktop Linux experience. It was a mess, hardware and software. The owners of the system decided it was cheaper to shut it down.
Like hardware, software also degrades over time⊠entropy in action. AI is software, AI also âhallucinatesâ.
How does AI fix an intermittent hardware error or complete failure? How would AI do a full root cause analysis of intermittent failures, hardware or software?
Users have other priorities that reflect the reasons they are there in the first place, large systems management is not why they were hired.
In other words, after 20+ years of hardware and software management of large sophisticated Linux clusters I would say that AI would never be capable of effective, solo, management. As an assistant for somewhat limited, general real-time support, fine. And since AI really is nothing but sophisticated, probability/statistic algoritms based on previous known situations, how could it handle an anomaly, a new, previously unseen software or hardware, permanent or intermittent, failure other than hallucinating? Or pull out a rack, remove a failed component and then replace it?
It will never replace solid, experienced System Administrators any time soon.
4 Likes
Very true! 
1 Like
AI cant fix hardware. I agree. It lives in hardware, therefore it cant fix it. That is like asking the brain to fix the body it depends on.
I see attempts to auto-manage software ⊠eg supervisor daemons⊠as admissions of failure. They ought to fix the software, not automate restarting it when it fails.
I think AI could do some software jobs ⊠the more pedestrian things like keeping versions of packages in line. There is nothing wrong with using AI to do some of the hack work. That does not imply that it could do everything.
I dont think Sysadmins need to worry about being replaced any time soon. The job might change though.
The other side of the story is I read recently about a server that ran for 16 years without human intervention, and only stopped then because of hardware failure. It is possible to get software bug free and requiring very little human attention.
3 Likes
Do not know where W11 got branded the âdifficult to useâ OS, it is pretty much automated, if one wants it that way, and a lot of userâs do!! Never have ran a mac, but running Linux without user input, I wish you luck!!
2 Likes
My last experience with Win was W10. It was easy enough to operate, but the update system was an annoyance⊠Nothing like as easy as Android, but not as choice-ridden as most Linuxes.
Some people say Mint is more user-friendly than Win, but most Linuxes are not, because they are for people who know what they want technically.
No idea about MacOS. My guess is it is better than all of them from a UI perspective.
Just think mint but with the maximise, minimise close in different colours and top left not right⊠If you visit control panel its almost the same. Just visual small changes. Told flisk between them you would pick up in a half hours no problems.
Its not hard to use, the issues come with updates and things not working after you are locked out during that and restarts. Most clients I see are about virus issues. Bet if I ran adwcleaner or malwarebytes on every windows user I see I would pick up a dozen issues within every one of them without fail !
Plus most try free AV or paid AV which slow users down and stop almost nothing.
Do you mean AV programs only clean up the mess after the event? âŠor worse , do they detect viruses and only inform the user rather than remove the virus?
I think the big problem is Win users download third party apps ⊠that is worse than using AUR or PPA.
Win needs to do something about how malware gains entry. Mopping up the damage is a bad approach.
1 Like
Yes, software. So it is only as good as it was programmed or trained on. Plus my understanding is that AI does not store any data except for a vocabulary, It has been âtrainedâ by feeding it huge amounts of data. AI once was arguing with me about Linux kernel 7.0 because the information was too new.
2 Likes
I believe the virus and Adware is not so much getting pass the software. It is getting by the user. Software can only do so much. It can not stop a user from downloading programs and letting them run on their PC, it can not stop user from clicking on links that are in emails or on web pages, it can not stop users from going to web pages that track them.
4 Likes
In a sense, all computer problems are human problems. A program does exactly what it was programmed to do. If a programmer writes faulty code, the program will execute that fault ⊠every time you run it.
In that sense, computers are 100% reliable (barring hardware faults). It is humans that need to be managed.
1 Like
In principal a AV program prevents virus attacks no matter what source. Most users think they are protected and download things infected as they dont scan or use a store (repository)
My tools are for cleaning after an attack not protection before.
That is another issue, for which I check using superantispyware. But also check any installed extensions and more recently notifications as that again causes popups not blocked by the tools.
Its getting harder and harder to spot or do, client last week had 50 objects picked up by adwcleaner, had to run it several times to get rid. It had been taking almost 1 hour to startup. Plus he had a strange antivirus tool he had paid for which only appeared to be making it worse.
1 Like
Can I see a show of hands of people who use AI? And if you use it, what was your reaction to it?
I use AI and I have seen where @Tech_JA and @Rosika have used AI in some of their posting.
So far I like AI. I mostly use Co-Pilot, but also use ChatGPT. They both seem to very good tools for retrieving information and perform a good job at problem solving.
Some thing I do not like about AI: they tend to give too long or too much information in a reply, they tend to tell you what is not the problem, and you have to be careful how you ask or word a question. Luckily since AI is a program, I have been able store or set some
preferences to modify the replies I get. One example is I told AI to give me short replies.
3 Likes
I use Grok and Claude mainly. I use AI when I am trying to learn or figure something out. It is of course not always correct, but it gives me a good starting point and vast amount of information in one spot to start looking at. There is definitely a learning curve on how to ask AI a question to get a more specific & accurate type of response. (Claude actually tells you - I am not always correct.-which is a nice reminder)
Not me, never used it dont want to. Everyone is talking about it here, took me a while to work it out as they kept saying Lea who I thought was a girl but they are actually saying lâIA (lâ because its a thing but identified as female as next letter is) intelligence (fĂ©minine) artificielle)
Being sexist why is intelligence féminine⊠I dont want to answer.
Iâm fully with you, bro. I prefer OI for any problem-solving.
1 Like
I use AI for some questions I have. Mostly technical like my DIY audio stuff: why my precious 45 tube is glowing purple? (And then I throw my ampâs schematic to AI.) Last time it spotted the culprit and my amp is working right. I had biased it way too hot.
Or like today when I was updating my system and I wondered why my system is pulling WebKit. AI gave me wrong cli commands when asking.
Most often AI just tries to âbe your friendâ and it is very frustrating. It tries hard to solve a problem in a way that it was just a small mistake or it tries to work around the real problem when all I need is it to say âThis is wrong, check this:â and give you an answer or a link.
I use duck.ai (worst) Gemini and chatgpt all free tier
4 Likes
I do.
At work I use Kiro (from AWS), Devin, and AWS DevOps Agent.
At home I use Copilot almost all the time. When I disagree with Copilot I check with Gemini. When coding I use the free versions of Kiro and GitHub Copilot in VS Code.
I have the same gripe. It could be something Copilot tends towards since we both use that, no idea really. I should do something like youâve done to have it give shorter answers.
At work when I have it walk me through something or troubleshoot something on AWS or in a cluster, it will give me many steps. I finally did tell it, âOne step at a time because many times an early step has an unexpected result.â In typical AI sycophancy it agreed with me that step by step was best.
3 Likes
I have found out thru questions to AI about how to set the preference of how I want Co-Pilot to respond to my inquires. These are my setting (AI calls it memory) for Co-Pilot. I would assume that other AIs have the same ability.
- shortâreply mode
- dry/blunt tone
- diagnosticâfirst workflow
- verificationâfirst workflow
- minimalâreasoning preference
- stepâbyâstep on command
Tech domain depth levels
- software: medium
- scripting: beginner
- virtualization: beginner
- backup: medium
- networking: beginner
- security: medium
- hardware: medium
- cloud: medium
1 Like
Why, if you donât mind sharing? I find AI is just a tool that is gives the same information available on the internet but is much faster. And you can ask follow up questions.
1 Like