It's a black box.
Magic goes in the one side, your new machine overlord comes out the other.

3b1b is always a great resource for wrapping your head around the concepts

Maybe take a class on it?

had a tab once about building a neural net from scratch but I lost it

Have you tried reading stuff or doing things yourself? I think videos are alright at getting basic information out there, especially as something in the corner of your eye, but as a primary source it seems quite terrible. Back in school the teacher talking would generally go in one ear and out the other while actually doing things would quickly teach me things that I would remember. Reading was also better for me and could give me the same information in about 10% of the time.
The really good 3D modeling tutorials, ones that people pay for/pirate or even see for free on youtube and the like often come with various materials including work files so that the stuff the guy is talking about is tangible to you and you can interact with it and do the same thing in the video with your own hands. I'm still not great at actually following through entire courses, but it's something that makes it far easier to commit things to memory.

>>119469
Doesn't machine learning require more processing power than the average person can reasonably pay for?
(I mean, to get to results that are worth a damn)

>>119469
I'm also one to learn more from practice than reading, but isn't reading/lectures the only way to understand the concepts of AI and the math required behind them? For example, I can program a neural network following tutorials and have a general grasp of what the functions do, what a loss function might be, what gradient descent means, etc, but I want to know what they mean in the pure mathematical sense as well.
>>119471
Don't know how it is now but a year ago when I was learning machine learning, you could train (albeit smaller models) with Google Colab for free.
Don't know how much the average person can reasonably pay for but I did also create a worthwhile image recognition program to categorizing my image hoards.

Yeah, I guess the big LLM stuff does require beefy hardware if you actually want to create competing products. People did mess around with making neural network stuff years ago before this chatgpt stuff took off, though, so I assume that stuff is still out there. Things like a guy training one to play Pong or whatever. While being completely ignorant of the inner workings of this stuff, I think that kind of thing is probably informative.

q-learning

>>119473
>I want to know what they mean in the pure mathematical sense as well
i asked a friend in the field the same question and was pointed to this lovely resource https://d2l.ai/
You will want to have a solid math foundation in areas like linear algebra, calculus, statistical methods, ect.. but it does also have some introductory chapters on the most relevant foundational items. I haven't made it past chapter 8 yet and have had been using it for a few years but I still think it's a great resource. It's almost exactly what I was asking for, as I was tired of half-baked medium articles and youtube videos that went which function to use to make a dog-cat classifier but they never told you much about how or why it worked.

>>119483
>d2l.ai
Nice, thanks, gonna get started with it.