The sheer difference in scale between the domain of “here are all the people in the world that have shared data publicly until now” and “here is the relatively tiny population of people being paid to add new information to an LLM” dooms the LLM to become outdated in an information hoarding society. So, the question in my mind is, “Why will people keep producing public information just for it to be devalued into LLMs?”
If you mean obfuscation, then yeah, maybe that makes sense to fit more into the window. But it’s easy to unobfuscate, usually.
Otherwise, I‘m not sure what the goal of an LLM specific language could be. Because I don’t feel most languages have been made purely to accommodate humans anyway, but they balance a lot of factors, like being true to the metal (like C) or functional purity (Haskell) or fault tolerance (Erlang). I‘m not sure what „being for LLMs“ could look like.
Even limiting yourself to code generation, there are going to be a lot of software developers employed to write or generate code examples and documentation just for AIs to ingest.
I think eventually AIs will begin coding in programming languages that are designed for AI to understand and work with and not for people to understand.