I have asked several LLM to output brainfuck and Malbolge, but none of them seem capable to do so, even those that implement train of thought, me included ofc.
To your first question, nop, I have no idea how much energy takes to index the web in a traditional way (e.g MapReduce). But I think, in recent years, it’s been pretty clear that training AI consumes more energy (so much that big corpo are investing in nuclear energy, I think there was an article about companies giving up meeting 2030 [or 2050?] carbon emission goals, couldn’t find it)
About the second… I agree with you, but I also think that the problem is much bigger and complex than that.