Could an LLM-only language cut tokens and coding costs?
The idea is to create a made specifically for LLMs, not for human comfort first. Each token would carry denser or more exact meaning, so an LLM could produce the same kind of code with fewer tokens than it needs in Python.
If enough existed, this might let LLMs write code faster and lower . A large could also hold much more useful work than it can with normal s.
Human-friendly syntax, such as semicolons or curly braces, may act like extra noise for an LLM, so removing that noise might improve . This is a theoretical proposal, not a proven result.
Key points
- A dedicated LLM could express code with fewer tokens.
- Fewer tokens could mean faster output and lower .
- A could fit more useful code and task information if the language is more compact.
- The idea assumes enough exists for models to learn the language well.
- This is speculative and has not been shown to work yet.