Are we sure GPT is going to improve 10x in 10 years? Hasn't it already been trained on the vast majority of available text data? We might get incremental improvements, but it's not like we have 10x more data lying around somewhere to feed GPT5.
10x? Try 1000x. We haven't even tried hooking up different types of reasoning models (chess engine, weather modeler) together yet...eventually LLM will be just one aspect selected based on intended use
GPT4 is basically Pong. Within a few years we will be nostalgic for its surprise value
HN has a God Complex when it comes to people like Carmack though, so you can't really survive disagreeing with him
I don’t have a very good sense of what kind of information that GPT-4 had access to, but I imagine there is a whole world of knowledge locked up in books, particularly textbooks, journals and periodicals that it doesn’t have access to. That could be interesting.
