Sanskrit is the best language for the LLMs.

Wait, let’s not conclude yet, there’s more than that.

English or any other language has a problem of ambiguity.

For example:

“I saw a man with binoculars”

Possible interpretations:
1. I used binoculars to see a man
2. I saw a man holding binoculars in his hand

Sanskrit doesn’t have these ambiguous issues, because of its structured language and logical syntax.

Panini’s Astadhyayi scientifically is a highly precise grammatical system, with a belief that it’s also a scientifically structured language which has no ambiguity.

It’s even a language that has the capability to express complex ideas concisely.

Well, that’s good for LLMs to become better right?

Yes, but there’s another problem.

Lack of adoption, cultural dependencies, it’s still a natural language and not practical.

Sanskrit is great! But, this is the reality.

Without widespread usage, there will be no real world dataset, and it’s not even spoken today.

LLMs aren’t just grammar, data matters a lot, that’s why English is currently a good fit due to its vast dataset.

LLMs need better disambiguous, context aware models, no doubt, but we cannot simply revive an ancient language just for LLMs.

Sanskrit might be theoretically ideal, but practically impossible.

Well, I leant in the process that there’s a language in making called “Lobjan”, will post about it more when I learn more about it.