Introducing LMLang, a language for LLMs by LLMs

Last weekend, I was thinking about several things and I might have drunk two or three beers, when the following thought popped into my head: computer languages are made around the limitations of humans to understand logic and code, LLMs have a different set of limitations and strengths, what would a language made explicitly for LLMs be like?

So, of course, I asked Gemini (my go to when I have idle thoughts) and Claude (my go to for coding stuff). They told me that it would indeed be different, a language for LLMs would need to keep in mind context and token cost, it would be easy to follow and clear, for an LLM, not necessarily for a human.

The next step was asking them to build it. I told Claude to come up with a description of this language, basically a blueprint to build it and then to make it clear that I didn’t want to be interrupted while watching cartoons and drinking beer I told it: “this is your project, buddy, don’t ask me anything, if you have any questions, ask Codex and Agy, try to get consensus and delegate the work. Only come to me if it involves giving plutonium to Libyan terrorists” (we often use Back to the Future references).

After a few hours (Claude asked Codex to do some heavy stuff using its top model and it hit the usage limit a few times, I told Claude to just wait each time because I’m not made of money) they came up with LMLang (I named it, probably should have spent more than half a second thinking about the name). It can transpile to Python or make binaries with Go. I did ask a few questions which resulted in a few features being added or changed, but the language is 90% theirs. They though about it, worked out its problems and design issues.

I haven’t tried extensively, at least nothing serious for now. So far I have just made a couple of simple text games and calculators, but I think it’s robust enough to attempt more serious work. If you want to give it a try, get it from https://github.com/rb3m/LMLang and tell your favorite LLM to use it. Feel free to fork it. However, remember that it’s a language for LLMs, so any contributions you should ask them to make them.