Interesting article about AI

[gidesa]

No, it isn't.  A good encyclopedia is verified for accuracy, AI will blurt _anything_, no matter how preposterous with the "confidence" of a world class expert.
AI is a digital politician, you always get an answer.  Correctness is not a concern.

Indeed the AI assistants used for programming are generative AI, that doesn't have any type of "score" in its answers.
Other type of AI, as image classification or object recognition neural networks, always return not one, but a group of answers, with their scores.
So you have a measure of certainty/uncertainty of that answers. 
Then many, if not all, AI tools always answer your question. The generative AI, as of today, cannot (or is not programmed to) return more than one single answer, without any certainty measure.
Of course it's more impressive for people  a single pseudo-true answer, that a list of answers with score.

[Bogen85]

To be arguing against AI programming tools is like arguing against encyclopedias,

No, it isn't.  A good encyclopedia is verified for accuracy, AI will blurt _anything_, no matter how preposterous with the "confidence" of a world class expert.

AI is a digital politician, you always get an answer.  Correctness is not a concern.

Owning a calculator (or an encyclopedia) does not make anyone a Mathematician.

I was making a point about using encyclopedias at home versus walking to the library.

Also, an AI saves me the time of reading a whole set (which may take weeks if that is all I did when I need to get problems solved now).

There are also some AI models (You.com's Research and OpenAI's Consensus) that are forced to cite all their references. They take a bit longer to answer (a minute or so versus under a minute), but what they are saying is backed up by human authors knowledgeable of those subjects.

I've also said if in a few minutes of a conversation I know I can't rely on what an AI is telling me that I'm not going to waste my time.

I know AI's can spew garbage answers. Like I've said, if I don't know like where an AI is going with something after a few minutes I'm not going to waste my time with that approach.

If those that use AI tools regularly took everything an AI said as accurate they have chosen the wrong tools to use and are obviously living in a fantasy world with no connection to the reality of getting real work done with computers.

[Bogen85]

There are also some AI models (You.com's Research and OpenAI's Consensus) that are forced to cite all their references. They take a bit longer to answer (a minute or so versus under a minute), but what they are saying is backed up by human authors knowledgeable of those subjects.

And I check the references. Occasionally the references are dubious, just like references of articles written by human authors.

[Bogen85]

No, it isn't.  A good encyclopedia is verified for accuracy, AI will blurt _anything_, no matter how preposterous with the "confidence" of a world class expert.
AI is a digital politician, you always get an answer.  Correctness is not a concern.

Indeed the AI assistants used for programming are generative AI, that doesn't have any type of "score" in its answers.
Other type of AI, as image classification or object recognition neural networks, always return not one, but a group of answers, with their scores.
So you have a measure of certainty/uncertainty of that answers. 
Then many, if not all, AI tools always answer your question. The generative AI, as of today, cannot (or is not programmed to) return more than one single answer, without any certainty measure.
Of course it's more impressive for people  a single pseudo-true answer, that a list of answers with score.

Not sure about "always answer your question". Yeah, it use to be a lot worse. The models are getting better about saying what you are asking for is possible.

Any more I don't straight into the solving the problem.

I first ask, no code please, I just want to know if using A with B can I do C.

If it says that is not possible, I ask for references as to why not possible and I do a quick sanity check elsewhere based on the the reasons why not.

Same for if it says it is possible. I ask for references from documentation, forum threads, articles, etc.

If I'm satisfied with that, I cautiously proceed.

But I'm not going to waste my time with garbage answers, I learned not to early on when I was figuring out what AI tools can do.

[Bogen85]

To be arguing against AI programming tools is like arguing against encyclopedias,

No, it isn't.  A good encyclopedia is verified for accuracy, AI will blurt _anything_, no matter how preposterous with the "confidence" of a world class expert.

AI is a digital politician, you always get an answer.  Correctness is not a concern.

Owning a calculator (or an encyclopedia) does not make anyone a Mathematician.

My taking advantage of AI tools is based on the foundation encyclopedias had in my learning early on.
Without the foundation encyclopedias gave me AI tools would not be able to assist me in research and problem solving.

Just like a calculator is not useful if you don't know math, but it is useful and saves you time if you do know math well.

[gidesa]

Not sure about "always answer your question". Yeah, it use to be a lot worse. The models are getting better about saying what you are asking for is possible.
Any more I don't straight into the solving the problem.
I first ask, no code please, I just want to know if using A with B can I do C.
If it says that is not possible, I ask for references as to why not possible and I do a quick sanity check elsewhere based on the the reasons why not.
Same for if it says it is possible. I ask for references from documentation, forum threads, articles, etc.
If I'm satisfied with that, I cautiously proceed.
But I'm not going to waste my time with garbage answers, I learned not to early on when I was figuring out what AI tools can do.

For example, models for object detection ALWAYS return a list of objects, between the ones that they were trained on. But with a score for everyone, so you can easily discard what doesn't satisfy your criterion. So, say, the model is trained on dogs, and the image doesn't contain dogs, then the model ALWAYS will return a list with dog, but with a score very very close to zero (although occasional "allucinations"/errors are always possible, and it will return an high score).
Interesting that some generative models now have a list of sources.
Of course the work to verify the references has to be far less than the work that you want done from AI! :-))

[Joanna from IRC]

Lainz yes I believe that we should all understand how assembly language works even if we program in pascal.

First of all I’d like to say that AI is not a good analogy for fpc versus assembly language. For one thing, fpc is open source, ai is not so far as I know. Fpc has real programmers that can be talked to and questioned. Does chat gpt or other ai projects have forums where you can ask the creators of it questions about how it’s written ? I’m not Interested in talking to AI but I’m very much interested in talking to the creators of AI.

Anyone who cannot explain something in layman’s terms is probably either up to no good or doesn’t understand it very well. To be honest I’m not interested in learning from any human who can’t answer my questions or even worse is too arrogant to admit to not knowing the answers and rudely tries to cover up his ignorance by accusing me of being too lazy to go read books.

I have doubts that any of the people who love AI have even read the article about it. Of course being spoon fed info from ai is easier than writing all the code yourself. I’m not going to argue with that. But what about the long term consequences of being dumbed down and becoming reliant upon the creators of AI and their plagiarized magic answers that they gathered up from everywhere they could find.

I know it seems counter intuitive but making things easier can and has lowered the quality of things and made them cheap and disposable. At no time in history has it been easier to build things than now. All this has led to is building an abundance of ugly energy wasting windowless boxes. There are buildings that were built better in the 1900s using mules.I don’t want software to become the equivalent of a monoculture of big box architecture.

There is always this promise of efficiency and people will have more free time.Yet somehow that’s not how it turns out is it? I’ve heard that medieval peasants had more free time and holidays than modern workers.

[Bogen85]

I did read the article. It did not have enough technical depth to argue for or against the merits of AI for me to have much of an opinion of it.

I've bought books that I've learned from that are not "open source".

I'm not free when it comes to me writing software for others.

I don't mind spending less than than $20 a month to a consultant (or two) that will save me days or weeks a month of my time helping me write my own code.

I still use open source programming languages to compile and build my code. Yes, my code. I own it, I understand everything that is being done in it. If I do not understand it, I can't own it, and it can't be mine. (Or my company's code when I do it for those who have hired me, and know I'm using AI to assist me). But even my company's code that I create for them, I must "fully own it" from a knowledge standpoint, and they need to "fully own it" from a legal standpoint.

If I don't completely understand the code an AI assistant helped me write, I don't care if it works of not. It would be useless to me. I don't even bother trying to compile or debug if it if I don't completely understand it. I'm far faster at reading code than I am at writing it.

Yes, there are the ongoing legal discussions on if the code was copied from existing source code. While that is an "interesting" debate, the inexpensive "consultants" assign the rights to those who ask them for assistance. I know others experience varies. In AI image generation (that I've found to be unreliable, maybe I've tried it wrong, for me I've for the most part lost interest in it) it does not appear to me that the image is being generated based on my criteria, but is rather some image altered to match my criteria. I guess because LLMs (Large Language Models) are good at translating and generating textual content in the language (natural or artificial) based on the what was asked to be generated or translated, and that is much more straight forward than image generation, and much more controllable by the one availing themselves of the consultant.

[Joanna from IRC]

I often listen to a podcast https://the-quash.captivate.fm/ wherein an attorney talks about his experiences talking to chatgpt. Apparently the answers that are given depend entirely upon the questions it is asked.
So a normal person who doesn’t have a lot of skill asking questions will not get the truth out of it. 

[cdbc]

Hi
Joanna, that's basically what everybody have been trying to tell you, here in this very forum!!!
Regards Benny

[Joanna from IRC]

I’ve often found that in order to ask good questions you have to be knowledgeable about the topic that you’re asking questions about or even know what the correct answer is already.

I once needed help making a complicated cte with sql for my project but only had rudimentary knowledge of sql language. I didn’t know all the proper terminology and jargon with which to ask the question and most of the people could not understand what I was talking about and tried to give me unhelpful canned responses. It was extremely frustrating for everyone involved. Then I went somewhere else and found someone who was more intuitive than the others who showed me how it’s done and gave me some code that I could modify.

I seriously doubt that AI can do better than live people at coming up with good answers. I don’t want people to be replaced with AI because it’s “faster”. That will be a as useful as only being allowed to talk to to a computer on the phone which puts me on hold and then hangs up. [Yes this is a tactic used by people who steal money}

Yes things are getting worse the more technology is used to replace people. Nobody even seems to care what will happen to hundreds of millions of people who could end up with no way to support themselves  because they were replaced by Ai.

[440bx]

Yes things are getting worse the more technology is used to replace people. Nobody even seems to care what will happen to hundreds of millions of people who could end up with no way to support themselves  because they were replaced by Ai.

That may be a valid argument but it is a precarious one.

The Luddites made that argument long ago and they were partially right.  Those jobs, basically no longer exist due to machines.

The ever present goal of today's societies (regardless of economic model) is to maximize productivity.  This is what lowers costs, something that you and just about everyone wants done (people are always trying to get the most for the least amount of money.)  Machines accomplish that at the cost of human jobs.

The important point is that, in the long run, AI or any other form of automation can only replace humans in the performance of an activity if and only if it becomes capable of performing the same activity to a somewhat acceptable level.

When it comes to programming, AI is barely capable of providing a programmer with pieces of code that can be pasted together.  Some self proclaimed "programmers" do have something to worry about but, for some reason I find it hard to sympathize with them.

Your comment reminds me of a conversation I had just a few days ago about a company replacing customer support personnel with AI.  Basically replacing very poorly trained people who regurgitate stuff written on a sheet of standard questions and answers.  These people are usually  in a call center in a third world country, usually barely speak the language and, they usually have such a heavy accent that they are difficult to comprehend.  These people are rather likely to be replaced by A.I

A.I can help automate some tasks as long as the task requires _no_ intelligence whatsoever (and even that is limited by the user's ability to ask questions "properly")

The day that thing can figure out a shorter (and correct) proof to Fermat's last theorem entirely by itself, that day there will be something to be concerned about.

Today, the concern is that people of limited capabilities are going to misuse A.I because they want to be part of the "in" thing.  Binary fashion, "this program is an Armani/Dior/Lanvin/Gucci/"whatever floats your boat".  Move over Knuth.

[Joanna from IRC]

This is where we start to delve into a philosophical discussion. Why do you think it’s so important to just strive for “productivity” at all costs. Historically there was competition over resources and those who advanced their population and technology the fastest had the best chance of winning wars and plundering everyone they could defeat.

What price are we willing to pay for increased productivity? Would it be ok to eliminate all superfluous members of the population? That would certainly increase productivity. What benefits does “increased productivity” is it really necessary to replace all possible jobs with substandard bots.

It’s a well known fact that modern life is not making the majority of the population happier and healthier. Many people are lonely/isolated and/or working long hours under stressful conditions. Endlessly pursuing materialistic goals. Modern so called food is Often full of weird substances which can hardly be pronounced.

I don’t think most people could ever benefit from stuffing AI Into everything possible. Based upon history it will benefit very few people who want to get rich by cutting costs. AI squanders resources which could be better used elsewhere.

[VisualLab]

The day that thing can figure out a shorter (and correct) proof to Fermat's last theorem entirely by itself, that day there will be something to be concerned about.

Exactly. And it probably won't end as positively as in Lem's book "Golem XIV" (this can be considered a positive ending for people).

[Joanna from IRC]

I think a lot of people don’t understand what programming is