Dalija Prasnikar

Yes, but people usually don't plagiarize every time they open their mouth. Comparing to AI which are trained mostly on copyrighted content, where this is way more likely. Just the other day when I was verifying something AI spit out whole, long, Stack Overflow answer which was already deleted and visible less than a week. This is the difference I am talking about. Again, when you get some AI output you have no idea whether that particular output is infringing someone's copyright and having such code in your codebase is a liability. If you have to write similar functionality, it will be your own work, unless you literally copy paste it from somewhere. And then you will know what you did.

This covers some similarities in predicting next word when listening to something. That is a completely different scenario from learning and reasoning.

If a human violates copyright it is called a plagiarism. You can actually sue for copyright violations.

Because humans don't have the ability to reproduce the content in a way AI can nor in the amounts AI can. Again the basic difference in how human learn and how the remember is why it is wrong to use AI. If you cannot see the difference I cannot make it clearer. It should be obvious. LLMs are statistical models, they don't learn anything. If they would be actually capable of learning then they wouldn't have problems with hallucinations or inaccurate information. Yes, their capabilities are impressive but they don't actually solve problems the way people do.

I don't expect too much protection from EU. Most of the time they have no clue what they are doing (Cookies). The only fair and proper protection would be explicit opt in for any AI training. Also it is very likely that some protection or compensations will be given based on scale of work. So some popular authors might even get some compensations, while majority of people will not be protected and will get noting out of it.

There is a whole world of difference between how humans learn and understand and how machine does it. when you truly understand some concept then you are no longer copying, you can write similar piece of code because that is how it is done. But you wouldn't be able to reproduce large chunks of some of my more specific and recognizable code. Even I couldn't do that. And you definitely wouldn't be able to reproduce all the things you ever read about. This is completely different from what AI does and how it can reproduce large parts of content it has been trained on.

I stand corrected. They settled. But I wouldn't call that a win in the context we are discussing. What they did was clearly wrong and all other companies are doing the same. They are training models on copyrighted data without permission. That is all I need to know.

Citation needed. There are ongoing lawsuits and result of those can go either way. So using potentially tainted code in your own codebase until there is legal precedent which makes it clear that such use is not infringing someone's copyright would be irresponsible at least. Anthropic has already lost one lawsuit when they trained on books. https://www.anthropiccopyrightsettlement.com/ so it is not unimaginable that they also violated copyrights with code training, too.

What attitude? Writing specifications and documentation is a good thing. Maybe it took AI for many to start doing that, but this is a still a good thing.

Eh... so we needed AI to rediscover value of writing specifications and documentation

That is not a modal dialog. That is dialog with callbacks - asynchronous dialog. Showing modal dialog is a blocking call and any code which follows will be run only after modal dialog closes and returns you result from ShowModal function. Showing asynchronous dialog (the Android kind) is not a blocking call. Which means any code written after that call will run immediately while the dialog is still active. And the logic you need to run after the dialog closes will have to be written in a callback method which is passed as a parameter to the ShowModal procedure. https://docwiki.embarcadero.com/RADStudio/Athens/en/Using_FireMonkey_Modal_Dialog_Boxes

In general if something is wrong with some post I would specifically mention the problem, but before AI era articles were written by people who genuinely wanted to share some knowledge and help others and invested some time in doing so. As more and more people are taking shortcuts and using AI to write for them because it is faster or whatever, then I am really not going to spend my time helping them and pointing to exact errors. I am not going to be an AI slave. I mentioned this here more to make others aware of incorrectness so they don't waste their time trying to learn something from that article, and to raise awareness that more and more articles are being written by AI and that many have not been properly verified. Some people might use AI to better rephrase or translate their own content and usually one can tell the difference between such usage and using AI to generate content. I don't have a problem with the former, but I don't approve of the latter.

I am not a huge fan of people using AI for writing knowledgebase articles. But if you really want to use AI for that, then the bare minimum you should do is to verify the correctness of what AI wrote. This article is full of nonsense.

This is my experience, too. It just keeps dying. Killing and restarting makes it work again for some short period of time.

I tested your code and I don't see too much difference between different variations, so it is harder to analyze what is going on. I have few suggestions, though. In many places you are inserting and deleting items from the position 0 in the list instead of adding it to the end. This is not optimal solution if there are multiple items in the list and it will require moving other items. You are also locking main thread to much in the GetUpdates function. You can easily check whether syncList.Count is greater than 0 outside the lock and then enter the lock only if this check passes. You would then have to check that again within the lock, but this would save you some time if there is nothing to process and you are calling that function frequently. Timer based version is also better than thread based one since you cannot overload the main thread with requests if it is too busy with some other work. I would also optimize DrawUserInterface method and cache some of the objects you are recreating there.