Winter is coming (AI-assisted coding and its influence on Free Pascal)

[dbannon]

That’s part of the divide and conquer strategy to split us all up into a bunch of squabbling groups with unsolvable differences.

No, not really, the "Media Barons" don't want to divide or unite, they just want us to look at their advertisements. They will show you absolutely anything as long as you look at their ads.

But getting back to cpicanco's question ?

First of all, several people have mentioned adding AI to the compiler, that, IMHO, does not make sense at all. Adding AI to the Lazarus IDE is arguably more interesting. A button we press and then describe the method we want written ?  Maybe.  But referring to my earlier post that only a small percentage of "programming" work is actually writing the initial code, who cares ?  The AI cannot help with the overall design, debugging or extending an existing application ?  Updating to suit new platforms or end user requirements ? 

Sorry, its not really going to be a big help and I might end up with some code in my project that I don't understand ? Or has a dubious copyright ?

I don't think so.

Davo

EDIT : And the title ?  I suspect the quote was "Winter is coming", George R Martin, no 'The".

[Joanna]

  No, not really, the "Media Barons" don't want to divide or unite, they just want us to look at their advertisements. They will show you absolutely anything as long as you look at their ads.

Be advised that the shows between the ads are also advertising ideas {propaganda} to you. 
In my humble opinion that is where the majority of influence happens.

[cpicanco]

EDIT : And the title ?  I suspect the quote was "Winter is coming", George R Martin, no 'The".

Thanks for the correction

PS.: Prefixing subject with "the" is an example of "portuguese thinking".

The AI cannot help with the overall design, debugging or extending an existing application ?  Updating to suit new platforms or end user requirements ?

As long as you have language dependent tasks, Language Models can help you (you may need some fine tuning). The more robust your own language is, the better the model will be. Also, currently, there is some sort of limiar and domain specif limitations. But limitations doesn't mean impossible. You must known enough stuff about your domain of interest and must setup a proper context to take advantage of language models.

So, my point is about tooling and is not off topic at all. It is a call for action. Maybe a concrete way to start is to decide how to train a model for fpc coding, designing.... What are the datasets we want? What are the tasks we want AI to assist? Then, we will need some sort of funding, because AI costs a lot. Maybe sending an email to someone at OpenAI? Who knows?

[dbannon]

Thanks for the correction

No, no correction. My comment was more to Mark who in a previous post miss credited the quote.

As long as you have language dependent tasks, Language Models can help you (you may need some fine tuning).

Certainly. But as I established, maybe help but not a lot.  But more importantly, at a cost -

Sorry, its not really going to be a big help and I might end up with some code in my project that I don't understand ? Or has a dubious copyright ?

I don't want code I cannot understand in my project and I most certainly don't want, as any cost, code in my project that breaches someone else's copyright.

Are you sure the help you got in your project is not subject to some unidentified copyright ?  I understand you are an Academic, you must publish your research and always attribute your sources ? 

There is a lot of Open Source code out there, code that the AI can spend it's quiet evenings reading. But being Open Source does not necessarily mean its free for the AI to take and give to you. Most Open Source code is subject to a specific license that may require attribution, may obligate the user to mark the work, it self, the same license, may put constraints on how it is used. Its even harder if the Open Source code does not have a license, then default conditions apply, they vary around the world and are vague and confusing.

Sorry cpicanco, this is sounding like a personal attack and its definitely not meant to be. I am sure your academic behaviour is totally correct. But if we let the AI contribute to our projects, we need to understand the implications. If you quote, as co-authors, every one who indirectly contributed to your next publication, it could look like a High Energy Physics paper.

So, your "Call for action" - please AI, tell us what we ask, but also tell us how you (the AI) found the answer.

Davo

[MarkMLl]

No, no correction. My comment was more to Mark who in a previous post miss credited the quote.

I didn't. I highlighted the subject line since the term "AI Winter" is well known to those of us "of a certain age".

I don't want code I cannot understand in my project and I most certainly don't want, as any cost, code in my project that breaches someone else's copyright.

Also we don't want code that the author can't work through logically and demonstrate its correctness.

An undergraduate should use algorithms broadly in line with what he's been taught, not something which could only have been written by somebody with many years experience of the tools and field.

MarkMLl

[Dzandaa]

Hi,

Interesting Links:
https://www.courthousenews.com/microsoft-and-github-ask-court-to-scrap-lawsuit-over-ai-powered-copilot/
https://techcrunch.com/2023/01/27/the-current-legal-cases-against-generative-ai-are-just-the-beginning/?guccounter=1
https://www.theverge.com/2022/11/8/23446821/microsoft-openai-github-copilot-class-action-lawsuit-ai-copyright-violation-training-data

Winter is here

B->

[Joanna]

I always knew that Microsoft taking over git hub was bad news. I would never put any code in a public repository.

This theft of people’s work on a grand scale will end badly I think. Who in their right mind wants to be exploited by for profit companies. Most people can’t afford to sue Microsoft and win. This will be the end of creative work in public domain.

[MarkMLl]

Frankly, I'm not that bothered either way. If I place something in any of the repositories I'm doing so because either because I've done something useful and feel others might be interested (e.g. an SpO2 sensor during COVID), or because I'm attempting to demonstrate that Pascal is still usable as a language (e.g. an ALSA interface). Or sometimes both (e.g. the drone interface software I'll be posting at some point).

And let's be honest about this: the GPL has now been reviewed and is considered enforceable, and there's lawyers around who'd love a shot at MS if they provably ignore it.

MarkMLl

[gidesa]

Hello to all.
Other than the "big giant" generative AI online (ChatGPT, Bard from Google, etc.) there are smaller nets than can run locally on a normal pc.
You download one of this  nets, around 4 GB. Then run a "simple" server executable on your pc, load the net, and have a local AI assistant. Interestingly,  these executables export a REST API, often with same syntax than ChatGPT api.
So it's possible to add some extensions to Lazarus (or Delphi) IDE that talk with the AI server.
For example 2 projects:
- gpt4all
https://gpt4all.io/index.html
- koboldcpp
https://github.com/LostRuins/koboldcpp

I think that koboldcpp is better.
An interesting use of these AI nets is to parse a natural language request/query and translate to some computer language, maybe Json, SQL, etc. It would be a great addin when a program interact with an user.
In the image an example, using koboldcpp,  with some commands and the translation/parsing in a Json format.
Note that you can write the request in many different mode (virtually all possible phrases with same meaning), the net is able to recognize and extract the data and fill the correct Json fields.
 

[paule32]

Hello,
I have started a Python ChatGPT toying Application, with it's help, I would provide simple Chat, and other things. Mainly I have started this Project for my private usa. But feel free, to drop a message, and follow the process of the little Tool.
The source code is currently holding in Python. But I think, it would be added Tools for fpc - I don't know.

Here is the https://github.com/paule32/ChatGPT_Template/tree/main to the Project.

Thanks for reading
paule32

[Handoko]

Recently, Google released a demo showing Gemini AI capabilities:
https://www.youtube.com/watch?v=UIZAiXYceBI

Unbelievable!

But if you try to find out more, you will know the demo was 'faked'.

[Joanna]

All this ai hype reminds me of this story about a guy who exchanged his soul for knowledge and worldly pleasures

https://en.wikipedia.org/wiki/Faust

[440bx]

The problem with A.I is that in spite of its being unable to derive even an infinitessimal understanding of the most trivial natural relations, many (too many) people will believe that what it "says" is (or must be) right and take actions based on that misguided belief.

A.I can be very useful to carry out tasks that require no intelligence whatsoever but, for anything that requires any intelligence, it's pretty much like handing razor blades to a rabid chimpanzee.  In the long run, it's not a good idea.

[Handoko]

Human brain is a result of million year evolution product. AI technology started in the 50 era, it still has a very long way to go.

[Joanna]

I once heard a story about ai learning how to land an airplane. It decided that the best way to land it was to crash it ....
It certainly is the best choice in terms of speed and fuel efficiency right?