3 ms·

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 persona

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.

awongh 2y ago
I thought I got a pretty good response- not o1: https://chatgpt.com/share/67db7ea5-f770-800e-bd6f-67945f584250 https://chatgpt.com/share/67db7ea5-f770-800e-bd6f-67945f5842... Maybe you meant that it immediately went to USA specific, but you can always change your question to clarify. Basically, if I'm unfamiliar with the topic this gives me a solid basis to ask begin to ask follow-up questions that use the terminology of the field, or start to find out what is regarded as important. I actually find it quite easy to go deep into a topic in a way that would be impossible in wikipedia, simply because if the topic becomes too dense for me it's hard to navigate your way back out- the wiki article begins to read like gibberish. With an LLM you can simply ask it to rephrase or clarify something.
- ants_everywhere 2y ago
  Yeah I mean that it went USA specific and also time specific. If you know enough to ask clarification that's great. But the problem with research is I don't know what I don't know, so when it puts me in a very small part of the problem space it's hard for me to see how little knowledge I've gained. In this example, I wouldn't personally consider any of these theories for why suburbs exist. They're all important reasons for the expansion of US suburbs during one time period. But that doesn't answer the question of why, for example, there aren't forces that pin everyone either to an urban or rural existence without suburbs. The answer to that question would involve rather different concepts than the answer it gave IMO.

awongh 2y ago

I thought I got a pretty good response- not o1: https://chatgpt.com/share/67db7ea5-f770-800e-bd6f-67945f584250 https://chatgpt.com/share/67db7ea5-f770-800e-bd6f-67945f5842... Maybe you meant that it immediately went to USA specific, but you can always change your question to clarify. Basically, if I'm unfamiliar with the topic this gives me a solid basis to ask begin to ask follow-up questions that use the terminology of the field, or start to find out what is regarded as important. I actually find it quite easy to go deep into a topic in a way that would be impossible in wikipedia, simply because if the topic becomes too dense for me it's hard to navigate your way back out- the wiki article begins to read like gibberish. With an LLM you can simply ask it to rephrase or clarify something.

ants_everywhere 2y ago
Yeah I mean that it went USA specific and also time specific. If you know enough to ask clarification that's great. But the problem with research is I don't know what I don't know, so when it puts me in a very small part of the problem space it's hard for me to see how little knowledge I've gained. In this example, I wouldn't personally consider any of these theories for why suburbs exist. They're all important reasons for the expansion of US suburbs during one time period. But that doesn't answer the question of why, for example, there aren't forces that pin everyone either to an urban or rural existence without suburbs. The answer to that question would involve rather different concepts than the answer it gave IMO.