The Year 2000 rule for IT services

Any service should operate perfectly on an infrastructure that was economically feasible in 2000. Is YouTube justified? Generally speaking, no. In general, modern services are excessively wasteful due to resource overload. So, it's considered justifiable to upload what should be podcasts and radio broadcasts as streams and videos. Moreover, today's resource overload encourages the creation of unnecessarily complex and expensive-to-produce content, at the expense of content that is truly justified by the creator's goals. In general, I'm a fan of ultra-low computing power and ultra-low memory. Things should be as cheap as possible. There's also the fact that expensive, cool things only look cool because people don't pay for them. For example, the Rabbit Hole blog, if hosted without any hacks, would be very expensive, comparable to the blogger's entire monthly income or even all of their assets, if you consider the lifetime cost of data transfer. And even with crutches, it will be a significant loss to the budget. Why haven't podcast services eaten into YouTube's niche? Have simple HTML pages not eaten up the walls of VK and LiveJournal? Because I was five years old in 2005, that's why. Hierarchy of Information Paul Graham's blog is a thousand times better than the average popular IT channel on Telegram. But it doesn't get 1,000 times more traffic, I guess, I don't know. But still. There are articles that are 1,000 or 1,000,000 times more important than others, but there's no structure to push them to the top of the search results. Why? The search is crap. Absolute, unimaginably bad crap. If I enter a word, I've already lost because I'm looking for a point on the map, not seeing the entire landscape. The recommendations are crap. Absolute, unimaginably bad crap. At best, recommendations simply feature content from other people I follow or who are part of a community. In general, a flat internet, where information is presented non-hierarchically, is a problem. Purely intellectual. Information has such a hierarchy. And it's not even a question of power. For example, a topic has subtopics; they exist due to the very structure of language. Why AI linguistic models haven't taken off. Because a word is a vector of intent, not just a piece of garbage in a giant graph of abstract concepts. Just because I'm interested in Paul Graham doesn't mean I'm interested in every blog by startup accelerator founder. My real intention is to read a few niche but influential people who have a lot to say about both information system architecture and some unusual things about society. On the other hand, creating a directory that only caters to my schizo interests is wrong. Short directories should rather create communities like LessWrong. But even that's inconvenient. That is, we need to somehow align the catalog with the user's intent. Perhaps generate it for each individual query. The Problem of Hidden Citations Lovecraft's influence is enormous. Billions of people have encountered his images. But he's not directly referenced, only through silly references. So, we can't simply create an h-index and shove top authors onto the main page. What should we do? Even if we create a perfect h-index for hidden influences, the search results page won't perfectly match the user's intent. That is, the search results must help process the token, not simply be relevant or, conversely, adequate to the structure of intellectual influences.