This page is still unfinished, but I think it still serves as a nice introduction about AI.
You can scroll down to skip the explanation and get to my opinions directly.

AI, What It Is, and What It Is Not

The Internet as a whole largely seems to misunderstand what AI is. AI, ML, LLMs, and generative AI are all slightly different things (and the acronyms don't help).
Artificial Intelligence (or AI) refers to an algorithm which attempts to mimic part of human intelligence. This can be achieved in countless different ways and has existed for a long time now, almost as early as computers themselves. The definition of AI is, in fact, so vague that truly anything that makes a decision somewhere based on human interaction could be considered "AI". I love AI based on that definition. Without it, you wouldn't be able to play against a computer in any video game, or have cool interactions with NPCs in story games.
However, this definition of Artificial Intelligence isn't what most people think of as AI (at least from 2022 onwards). Instead, most of them are referring to generative AI.

Before getting to generative AI though, there's an important distinction to make between (what I call) a static AI (also referred to as a reactive machine), and a learning AI (also referred to as limited memory). A static AI has been programmed in a certain way by a human and is, in the context of games, what most video game players used to refer to as "AI" before the AI boom in 2022. A learning AI, on the other hand, uses training data to learn what the best decisions are in a given context for its specific problem. The distinction between the two can sometimes get muddy if a static AI is programmed to "remember" previous decisions made by the human agent, but generally a learning AI involves some Machine Learning (ML) tools, which almost always involves Neural Networks.

Finally, generative AI is a learning AI which (generally) uses multi-layered neural networks (deep learning) to generate "something" (typically words or images) from human readable input. The goal of generative AI is to mimic its training data in its output. The difficulty of this task comes from mimicking output which it might never have encountered in its training data; yet, numerous Large Language Models (LLMs) and visual models now manage to cope with novel inputs incredibly well, at least insofar as making it "look" correct. Still, being able to generate such consistent and natural speech and images is a marvel of engineering which I still don't fully understand and that we too easily take for granted.

So, to recap, there's AI, which has two main branches: static, and learning AI. The learning AI models use Machine Learning and training data. Some learning AI models are generative AI and use Deep Learning (big Neural Networks) to generate content. A subset of generative AIs are LLMs, which are trained to mimic human speech.

My Personal Opinions on Generative AI

The problem has never been with AI. AI as a whole is very vast and has many academic branches which have yet to be explored. Even learning AI models are not problematic and can be extremely useful (they were already used in various fields successfully before 2022). The contreversy lies in generative AI.

I want to expand on this section further when I can (TODO) and writing down everything is very difficult, but the fundamental issue is this:

Generative AI models use training data, but are actually incapable of reasoning on that training data (unlike a human).

This means that (in the best case scenario!) they can speed up tasks that are present in their training data. In practice, this means it can be relatively useful at helping you do something you have no experience in; but if you have any kind of "training data" of your own (i.e., you have experience in that subject), the best use case of AI for you is brainstorming (and similar tasks, such as getting ideas for debugging, which really is a sort of brainstorming). It will likely be faster, more useful, more accurate, and overall better for you to do it yourself beyond that. If you need something done quickly and you don't know how to do it, then sure, use AI to do a quick and dirty version of it, but for every other use case, consider very carefully whether it is actually better (in whatever sense of the word is needed for your situation) instead of just doing it yourself.

A generative AI model is not designed to do whatever tasks we want it to do, it is designed to mimic its training data. It has become so good at doing so that it even manages to deal with tasks that it's already seen examples of (or similar examples of) in its training data. But the best it will ever be able to do is only as good as its training data (and, I'm repeating myself, that's the best case scenario). Until an AI model can learn and improve upon its training data, it will be incapable of dealing with mistakes in its training data, performing tasks it has no examples of, or just reason on the data like a person would be able to. So instead of using geneative AI and potentially getting things wrong, why not use (or find) your own training data on the internet and make your own judgement and reasoning on it?
Add all the privacy, legality, and copyright issues, and you end up with a bomb which looks helpful (and sometimes actually is) but that can explode at any moment.
I urge people to be extremely careful of generative AI. You shouldn't trust the information an LLM gives you any more than you would an uninformed friend who knows the whole dictionary by heart but doesn't understand any of it, because that's essentially what an LLM is doing, except not with a dictionary, but with all the (right and wrong) information on the internet.