AI usefulness discussion

[Bogen85]

A few years ago I took a deep dive into exploring the usefulness of AI tools.

Wasted a bunch of time but eventually figured out what kind of stuff works and what does not.
And figured out how to quickly see the answers I was getting were or were not going to help me.

The whole "well, I got 50 lines of code and I then I spend hours debugging it".

No, simply no. If the tool can't fix it after one or two error messages and fixes, done. Abandon that line of problem solving, do it myself with research, maybe try some other approaches with the AI tools, but do obviously do something else productive rather than trying to debug code that will never work or that will give me the answers I was looking.

But even if I'm not going to use the code it gave (or even attempt to make it work), I can look at it, and say "Oh, I get the gist, I know what to research now so I can do it myself" (and the "do it myself" may or may not rely on any more assistance from AI tools).

[LV]

For a wide examination of AI in mathematics that is very readable, let me suggest "The Unreasonable Ineffectiveness of AI for Math" here:   
https://synthesis.ai/2024/02/13/the-unreasonable-ineffectiveness-of-ai-for-math/ . 

Thank you, I enjoyed reading this article

[VisualLab]

Google uses AI to write 25% of new code:

https://www.tomshardware.com/tech-industry/artificial-intelligence/google-now-uses-ai-to-write-25-percent-of-its-new-code-alphabet-ceo-sundar-pichai-underlines-the-companys-role-in-the-ai-industry-amidst-strong-q3-24-financials

They don't say how much of their code was generated before, so it might just be they added a bit AI to existing codegenerators.

That said, I do occasionally see people use voice assistant on their mobiles (and it makes some sense giving the horrible input device options), and of course x86 processors branchprediction and Nvidia's dlss is also AI based. (though that seems to be more a classical neural network than large sparse correlation matrices)

The opinions of the commentators posted under this article are interesting. In fact, they are moderately skeptical. The first thought that came to my mind while reading this article was: what does the author of the article mean by "new code"? How is "new code" defined according to him? The commentators reacted exactly the same way. And that's because the information from Google is just a statement, without any data that would allow verifying their notification.

[Curt Carpenter]

Don't forget that even if it is their profession to deny any responsibility with such kind of things that you talked about but in the end there must be someone that can be held accountable (obvious reasons).

And on a general note: Note that we did not even got deeper into things such as equality, discrimination, accountability and data
acquisition and manipulation which are a real treat.

But, I have to guess: Nobody seem to care about the why, only the how 

The view that "there must be someone that can be held accountable" is interesting and it would be worth exploring, along with the "why" problem.  Lots of links to the best and ancient philosophical questions.  But this is probably not the place.  Good point though.

[Joanna from IRC]

There is hardly any accountability now when it comes to software controlled things killing people. The people responsible suffer no consequences and even if they did some lowly flunky following orders would be the scapegoat. I don’t think entrusting code writing to ai is a good idea at all.

I saw this funny cartoon in chat
https://usercontent.irccloud-cdn.com/file/4CiR1jCQ/4AA2287C-7FEE-4C8D-A296-F42CD3DC2434.png

[Bogen85]

I don’t think entrusting code writing to ai is a good idea at all.

What are code reviews?

For any code that is to be trusted for anything, other programmers (multiple) need to analyze every line of code written by another programmer and see what is being done and verify that it is correct.

Same goes for every line of code written by an AI tool.

People write bad code a lot too and people sometimes get killed as a result. I've not heard anyone advocate for humans to not be allowed to write code when things like that have happened.

[Bogen85]

I don’t think entrusting code writing to ai is a good idea at all.

AI tools can find problems in code that are hard for humans to find.

Hate on AI programming all you want, many find them useful.

I'm not pushing for people to use that, that is a choice, I'm just say quit thinking that everyone that bets benefit from such tools are clueless idiots that want to show off, and that have not idea how the code works that was written with or with the assistant of an AI tool.

[Joanna from IRC]

@bogen85 you and your intentions of what to do with code and AI might not be true for everyone.

AI tools can find problems in code that are hard for humans to find.

ok let’s assume that’s true. Are you familiar with risk compensation? Yes lazy people who are not you will decide hey no need to code carefully anymore. “AI will do my work for me and catch all the errors”. What if AI doesn’t find the errors? AI doesn’t care about bad consequences of code failing. Yes it costs more to have humans review each other’s code but it’s better than some of the alternatives.

I'm not pushing for people to use that, that is a choice

I agree. I could care less if someone plays with AI without me. What I don’t appreciate are people repeatedly showing up here and demanding that AI be incorporated into Lazarus IDE no matter how many times people here object, these people keep coming back and trying to shove AI down our throats. 

People write bad code a lot too and people sometimes get killed as a result. I've not heard anyone advocate for humans to not be allowed to write code when things like that have happened.

That is a complicated issue. There are definitely things that could be done differently so that the same results could be accomplished much more safely. The safety of things should not depend upon programmers writing perfect code {As if that is even possible...}

[Bogen85]

I'm not pushing for people to use that, that is a choice

I agree. I could care less if someone plays with AI without me. What I don’t appreciate are people repeatedly showing up here and demanding that AI be incorporated into Lazarus IDE no matter how many times people here object, these people keep coming back and trying to shove AI down our throats. 

In every IDE I'm aware of (unless some new IDE is AI at it's core) only offer AI as an optional plugin, and it is not hard coded in.

I don't see how that would be any different in Lazarus.

I'm an advocate for AI tools for those that already find them useful and want to use them. Trying to convince others to embrace them for their own use would be pointless venture to me.

For reasons I've already mentioned (accidentally leaking IP) I likely would never avail myself of an AI plugin that would allow for that, in any text editor (IDE or otherwise) that I use.

And since I use the same the same text editors (or IDEs) for hobby programming as I do at work, I see no point for me personally to getting use to (or worse, dependent on) such an AI plugin at home.

[Joanna from IRC]

I’m curious does ai require connection to internet?

[Martin_fr]

I’m curious does ai require connection to internet?

No.

Well, you likely have to download it first. Not sure you can order it on DVD (via phone, without internet).

[Handoko]

I’m curious does ai require connection to internet?

No.

I use Blender 3D. Recent versions of Blender 3D have AI denoising feature. If anyone doesn't know what denoising is, read here:
https://www.youtube.com/watch?v=8PpFaR3mD3g
https://docs.blender.org/manual/en/2.92/render/layers/denoising.html

Previously it usually took me more than 3 hours to get the acceptable-noiseless render results on my i3 Gen6 with integrated Intel VGA computer. Enabling the AI denoising, I now can get similar noiseless render results just about 5 minutes. Take a calculator and do the calculation yourself you will know how useful AI is.

AI does not only useful, it also generates many problems. Mass unemployment, copyright issues, low quality AI generated content, etc you can easily find them on the Internet. Simply saying AI is good or bad, is not fair. One of the great thing about AI is, it is used in cancer screening and the accuracy is very high.

https://www.youtube.com/watch?v=3EV6ryG2j7E

[Joanna from IRC]

I would like to know how AI actually works. Is it opensource?

[440bx]

I would like to know how AI actually works. Is it opensource?

A.I is a bit like saying "a compiler", it's not specific, it's a characteristic and it can be implemented in a number of ways (though, I believe the options are currently limited... that comment is based on my very limited knowledge of implementations.)

Therefore, A.I isn't and, cannot be, open source.  A specific implementation could be open sourced, just like a specific implementation of Pascal can be open source.  Also, just like a Pascal compiler can be implemented in different ways, there are different ways of implementing an A.I engine.

[Joanna from IRC]

@440bx have you ever seen any opensource ai projects?