How do you imagine the universal programming language of future LLMs will look? I mean a language built by LLMs for LLMs, not for humans to read or modify. A language that can be used to build any kind of application: backend, mobile, desktop, web, embedded, etc.
I suspect it will not be a language in the traditional sense. Instead of describing how to build a system, it would describe the system's entities, relationships, goals, constraints, NFRs etc...
The application model becomes the source of truth, and the code will be the generated artifact. A single spec defined and code generated will be optimized for web, mobile, desktop, api, etc...
The spec itself can be in a yaml type format or a hierarchical knowledge graph...
A parallel question: do you think Python being so entrenched in LLMs model weights means that future new programming languages are just going to resemble Python? Corpus means models that write that language means more corpus of that language.
OH, but LLMs are even better at natural language than they are with Python (which is why you have to often go from PL code to NL to get it to do something well, then go back to PL from the updated NL). So any kind of new language they develop will definitely be some kind of natural language where they derive semantic meaning from it to quickly use some deterministic PL to do something.
I think that LLM Agents will communicate with each other using internal embeddings - the natural language LLM thinking on. The final output may be a platform-specific assembly or binary output.
When specifically asked, the LLM Agent may present the architecture and structure in human-redable language or on diagrams.
I do not see the real reason for LLMs to support so many programming languages created by humans and for humans.
The spec itself can be in a yaml type format or a hierarchical knowledge graph...
reply