An esolang idea that might (?) be fun to kick around. The idea is to take advantage of the fact that LLMs are somewhat resilient in their inputs (can ignore typos, infer intent, etc.) to have a multi-level language. Here’s a sketch, but I’m not wedded to any of it: Programs are UTF-8. Each input code point is interpreted as an opcode. (Many opcodes are no-ops, probably most non-ASCII code points.) Some opcode(s) cause LLM execution of the program text. The fun is writing a program whose opcodes do something useful, while still being natural-language-interpretable enough to use the second-level interpretation of the language as well. (For the no-LLMs crowd, a similar idea is to have a pixel-based programming language…but the input is always pixels from rendering a textual program. Then you can execute either the pixels, or the text from which the pixels result, and bounce between the two.)