I don’t think anyone has any doubts about Jolla engineers being able to use AI tools, understand exactly what they want it to do, evaluate and test properly.
6 Likes
Very nicely said, Sir!
Right but the point is that they would not be able to finish it without AI or maybe that is just bluntly said but we would not get the new core this fast.
IF those vibecoders don’t really know, what their AI toll produce, you can have mor bags, backdoors, potential exploits as result. I see this at that moment as not acceptable for use thta on my daily driver devices. And yes, I can’t stopp it. But that is not the question here 
1 Like
No, that is not correct.
An AI model consists of two parts - we’ll call them the ‘structure’ which contains the rules, the logic and the reasoning (i.e. the inference engine) and the ‘learning’ which is the data that the AI model is fed to make its deductions.
The structure is simply a computer program, a very large and complex one, but a computer program nonetheless. Computer programs are algorithmic, put simply a set of instructions which a CPU loads and follows. Instructional algorithms cannot, by definition, produce a random output. The output may be incorrect or unexpected (the definition of a bug?), but it is always predictable, precise and never random. Even so called random number generators or functions (e.g x = rand (y)) in computers are not actually random or unpredictable. They just use data sets that are so large that the output appears random. I seem to remember that Basic used the calculation of Pi to several thousand decimal places and then used the time in seconds since 1st January 1900 as a parameter to pick the ‘random’ number from the Pi progression. In other words if you knew the algorithm you could predict the ‘random’ number.
If there is a bug in the structure then you might get an unexpected or erroneous result (but not random, imprecise or unpredictable if you knew the algorithm with the bug) - but that is a human coding the structure wrongly, and no different to any other coding error in any other computer program. It has nothing to do with AI per se. Of course this happens with AI models, but it is not the ‘fault’ of the model, but the humans coding it.
Far more likely is incorrect or inappropriate learning - but even then these are not random or unpredictable results but the inevitable consequence of ‘Garbage In, Garbage Out’.
As a useful example I am currently working with a large legal services provider on an AI model to help judges, barristers and solicitors speed up casework research. If that AI model was trained by scouring the internet it would come up with all sorts of ludicrous conspiracy theories and present them as ‘random’ legal facts. So, this model doesn’t use the internet. Instead it is a ‘closed’ model, trained only on (in the UK) statutes that have received Royal Assent (i.e. they are established laws) and court reports from common law cases tried by a judge and/or jury. That way the AI model’s inference engine only has access to established legal facts, tried cases resulting in precedent law, and accepted legal arguments.
So, there is no random, imprecise or unpredictable output from an AI model that is the fault of the AI model, just poor human interaction as there might be with interfacing on whatever level to any other computer program.
8 Likes
Absolutely agree, but that is a problem with the human using the tool, not the tool itself.
1 Like
They can build a wole linux based OS around qt5. You think they cannot port it to qt6? There are other reasons.
AI:
Qt 5 offers both commercial and open-source licenses, while Qt 6 has transitioned to a subscription model for commercial use, with LGPL and GPL options available for open-source. This change means that Qt 6 no longer provides perpetual commercial licenses, focusing instead on subscription-based access.
Human:
I could see why qt5 is still being used.
2 Likes
Effectively they are - and this is my benchmark for this argument. True randomness or not does not change this meaningfully.
Yes; they are random and imprecise. Do the same thing twice and you’ll get different results. There is actively introduced randomness, called temperature.
Nobody runs with zero as the perceived productivity falls off a cliff.
Poor training data definitely is a thing too - and substituting verifiable correctness by likelihood.
1 Like
I’m sorry but this is a pretty pointless question, software development is incredibly varied and whether AI makes sense for any or all part of it depends on a million variables, including what kind of software and what kind of AI we’re talking about. I’d guess what you mean is “vibe coding” but that’s pretty much exclusively used by obvious slops, scams, and clickbait-y YouTubers.
5 Likes
I don’t think I have anything more to add to the discussion, just my opinion is that adding this badge or whatever to signal that it is developed with or by AI would lead to some unnecessary stigmatization. I am fine if someone provides the source code for the app developed using AI with commits reasonable to follow the thought process (and not 5000 added lines in single commit - looking at rootelegram).
1 Like
Extremly binary question.
AI: If it is used by a knowledgeable developer to fasten the developement and s/he reads all the code created and checks if it is fine, all is fine.
If it is used for vibe coding by non-developers who can not actually meassure, control and classify what is written by AI or refuse to read the produced code, I am against the use of it.
So a better question would be:
Are you against vibe coded apps?
Here I say yes, I am against that.
As a tool in the right hands? All fine.
Should I answer this poll as Yes or No?
Best regards
Fuchur
5 Likes
Of course, but that doesn’t mean its random does it?
Lets’s take your benchmark of my example of a computer random number generator, x= rand(y). If you call the function twice you’ll get two different results. Well that is stark staringly obvious because the time in seconds used as a parameter to pick the number from the Pi progression has changed between the first and second function calls. If you call it a third time, a day later, you’ll get a different result again - but if you think this is truly random you’re being deceived. If you knew the value of the time parameter, that the Pi progression was being used as a data set, and the algorithm, then you could work out what any ‘random’ number would be at any given point in time.
That’s hardly random is it?
Your quote of the LLM ‘Temperature’ (which for those that don’t know is not the ambient temperature measured by a thermometer, but a ‘coined term’ used to represent a parameter determining (oops, that word again!) how an LLM creates its inferences between highly creative at the top end and more conservative at the bottom end, usually in values between 0 and 1), disproves your own point. If you knew the value of the ‘Temperature’, along with knowing how its inference algorithms worked, then you could predict the LLM’s output - not random at all.
3 Likes
Your entire argument hinges on that random number generators are not truly random.
For the problem at hand this does not matter - to the human observer they are sufficiently random.
Same with slopotrons. With a given nonzero temperature, you cannot predict the output without also predicting the random number generator. No compiler or other productivity booster tool of yesteryear has (pseudo-)randomness deliberately built in even remotely to that extent.
You just keep going - you’ll convince yourself eventually 
2 Likes
Ideas - not people, remember?
Address my argument, please.
1 Like
Yes and no is my answer, you have to understand something.
If for example i vibe coded something and the first thing i said here was:
Heiiii gyzz i kodes somethinz amazbent
yust go here to tryy it: http://127.0.0.1/myAwwsumApp
i hope someone tells me some harsh words.
Ok, I’ll bite. What argument?
So … what?
How does that delegitimize the use of AI in coding?
If LLMs can produce ‘pseudo random’ output its only because a human has chosen to make it so (e.g. to introduce a ‘temperature’ variable), just like a human can choose to create a ‘pseudo computer random number generator’, or introduce a pseudo-random element in any other computer program (e.g. a slot machine game).
Whichever way you look at it, and unless you choose to ignore facts, computer programs, whether they are AI models or Sailfish GUIs cannot, of their own accord choose to produce random, or unpredictable results when a human has chosen the algorithm that created those results. Their output is always driven by human input.
I honestly do not understand why you are arguing against this - its pretty basic stuff
3 Likes
With estimated annual consumption of 9.1 trillion litres of water (equivalent of domestic water usage of 1.3 billion people) and 945 TWh of electricity by 2030 this “AI” thing is in pole position of the race to pollute and fry this sh*thole the world has become. Let’s keep blaming cow farts for it, though. But my personal choice, for that reason and a few others, is to stay away from it for whatever I can (happily) use my brain for.
13 Likes
Good for you, sadly I cannot because of the nature of my work and the company I work for enjoys the new shiny toys and AI is the popular buzzword and if you do not ride this wave then you are out no matter how useful or useless it is to some people.
1 Like
Maybe that’s why I have always been a freelancer…