How Generative AIs Are Actually Trained

I’ve been playing around with tools like ChatGPT and MidJourney, but I’ve always wondered how these AIs actually “learn” to generate content. From what I’ve read, it’s about feeding them huge amounts of text or images and having them predict what comes next, but that still feels really abstract to me.

I tried reading a bit about neural networks and training datasets, but it’s hard to picture how it all turns into something that can write essays or create art. I’m curious if anyone here has hands-on experience with AI training or fine-tuning models.

Is it mostly about trial and error, or is there a systematic way pros train these models? Would love to hear how it really works from someone who’s done it.