I use LLMs quite a bit (as you can see, for example, on my stream). I also consume a lot of podcasts, YouTube videos, Wikipedia articles, and books.
My personal take is LLMs and podcasts are not great for research. Wikipedia is generally outstanding for science and related topics, although it's clearly astroturfed on topics about companies and some political topics. Podcasts typically cover things very shallowly but can sometimes be useful if there's a topic well served by the interview-a-day format. YouTube has a lot of treasure but you need to be able to search and not get distracted by clickbait.
LLMs function exceptionally as next generation search engines. But their knowledge for research bottoms out very quickly unless you know a lot about the subject and are able to prompt it effectively.
For example, I asked ChatGPT o1 your question about suburbs and its answer was basically total nonsense. The third theory it mentions was white flight and the post WWII era, which can't possibly explain why suburbs exist internationally and have since ancient times. There's probably a better prompt where you define what a suburb is and why you're interested in it and otherwise give it some good parameters and it gives you some more relevant information.
BTW I more or less agree that foundational texts can be very important to read. Not always, of course, but often enough that if you read a lot you should consider it. Often I find they bring considerable clarity to a subject.