3 ms·

Reddit is mostly trash now, but here's the thing though: If people stop talking to each other, what are all the AIs going to train on? Like say a hot new game

Reddit is mostly trash now, but here's the thing though: If people stop talking to each other, what are all the AIs going to train on? Like say a hot new game comes out tomorrow, SuperDuperBuster (don't steal this name). I fire up Chatgrokini or whatever AI's gonna be out in the next few days and ask it about SuperDuperBuster. So does everyone else. Where would the AI get its information from? Web search? It'll only know what the company wants people to know. At best it might see some walkthrough videos on YouTube, but that's gonna be heavily gated by Google. When ChatGPT 5 came out, I asked it about the new improvements: it said 5 was a hypothetical version that didn't exist. It didn't even know about itself. Claude still insists iOS 26 isn't out yet and gives outdated APIs from iOS 18 etc.

Theodores 1y ago
I think you need to answer this by looking from the other end of the telescope. What if you are the developer of SuperDuperBuster? (sorry, name stolen...) If so, then you would have more than just the product, you would have a website, social media presence and some reviews solicited for launch. Assuming a continually trained AI, the AI would just scrape the web and 'learn' about SuperDuperBuster in the normal way. Of course, you would have the website marked up for not just SEO but LLM optimised, which is a slightly different skill. You could also ask 'ChatGPT67' to check the website out and to summarise it, thereby not having to wait for the default search. Now, SuperDuperBuster is easy to loft into the world of LLMs. What is going to be a lot harder is a history topic where your new insight changes how we understand the world. With science, there is always the peer reviewed scientific paper, but with history there isn't the scientific publishing route, and, unless you have a book to sell (with ISBN number), then you are not going to get as far as being in Wikipedia. However, a hallucinating LLM, already sickened by gorging on Reddit, might just be able to slurp it all up.

Theodores 1y ago

I think you need to answer this by looking from the other end of the telescope. What if you are the developer of SuperDuperBuster? (sorry, name stolen...) If so, then you would have more than just the product, you would have a website, social media presence and some reviews solicited for launch. Assuming a continually trained AI, the AI would just scrape the web and 'learn' about SuperDuperBuster in the normal way. Of course, you would have the website marked up for not just SEO but LLM optimised, which is a slightly different skill. You could also ask 'ChatGPT67' to check the website out and to summarise it, thereby not having to wait for the default search. Now, SuperDuperBuster is easy to loft into the world of LLMs. What is going to be a lot harder is a history topic where your new insight changes how we understand the world. With science, there is always the peer reviewed scientific paper, but with history there isn't the scientific publishing route, and, unless you have a book to sell (with ISBN number), then you are not going to get as far as being in Wikipedia. However, a hallucinating LLM, already sickened by gorging on Reddit, might just be able to slurp it all up.