Reflections from Infinity
This is the final post in a series on ಅಕ್ಷರ ಮಂಟಪ (Akshara-Mantapa), my Kannada implementation of Borges’ Library of Babel. We’ve walked through philosophy, linguistics, mathematics, and engineering. Now: what did I actually learn from all this?
This post is text-only, by design.
The Series:
೧. The Dream of Ananta: why infinity, why Kannada
੨. Exploring the Infinite Libraries: a hands-on guide to both libraries
೩. Taming Infinity: the mathematics of bijection
೪. Engineering Infinity: Rust, WASM, and making it run
೫. Reflections from Infinity: You are here
I. Finishing the Unfinishable
There’s a strange feeling that comes with ‘completing’ a project about infinity. The library contains every possible 400-character Kannada text. It has always contained them. What still surprises me is how mathematics can define these vast, precise spaces and give us, at least in principle, unbounded access to them.
Borges understood this. His librarians didn’t create the library; they wandered through it, searching for meaning in an ocean of noise. In that sense, I don’t feel like a creator so much as someone who found a way in, and left a door open behind me.
And yet. The code is pushed. The site is live. The blog posts are written. By any reasonable measure, the project is done. But infinity doesn’t care about reasonable measures, and neither does the part of my brain that keeps thinking about weighted distributions and language-like randomness.
ಮುಗಿಯದ ಕೆಲಸ. Unfinished work. Maybe that’s the natural state of anything touching the infinite.
II. The Sadness of Noise
I wrote in Post 3 about why most pages look like gibberish: uniform distribution versus Zipf’s law, the 2,000x gap between how often common graphemes appear in natural Kannada versus in the library. I won’t repeat the mathematics here.
There’s a melancholy to it. The library contains every Kannada poem that will ever be written, every insight, every prayer, every confession. But it also contains 57,324^400 pages of static for every page of meaning, and there’s nothing on the page itself to tell you which is which. You only recognize meaning if you arrive already knowing what you’re looking for.
III. A Future Dream: Weighted Infinity
Sometime, when there’s more time and motivation, I’d like to build a variant with weighted distribution. What if common Kannada syllables appeared more often? What if the randomness followed linguistic patterns, not uniform chance?
This would break the purity of true infinity. Weighted distribution means some pages become more likely than others, which means you’re no longer covering all of possibility space equally. You’re making choices about what “looks like” Kannada, imposing structure on the structureless.
But it’s an interesting problem. You’d essentially be building a rule-based language model, one probability tweak at a time. Not a neural network, not trained on data, but a hand-crafted distribution that captures something about how Kannada sounds and flows. A proto-LLM made of nothing but frequency tables.
Would the pages become readable? Probably not. Zipf’s law is necessary for language-likeness but not sufficient. You’d also need syntactic structure, semantic coherence, things that can’t be captured by symbol frequencies alone. But the pages might feel less alien. The noise might have a familiar texture.
Maybe someday. For now, I’ve kept the library pure.
IV. What Did I Gain?
A tough problem, solved. That specific satisfaction of wrestling with something difficult and coming out the other side with working code. The Babel Trilemma, the modular arithmetic, the grapheme segmentation, the dead conjunct problem. Each was a small puzzle that clicked into place.
Perspective on infinity. It’s easy to say “infinite” and move on. Building something that actually touches infinity, that enumerates a space too large to comprehend, forces you to sit with what that means. Most of everything is nothing. Meaning is vanishingly rare. The universe doesn’t owe us patterns.
The joy of building something for fun. No client, no deadline, no requirements document. Just a fascination with Borges and a love for ಕನ್ನಡ and a few months of evenings spent in the rabbit hole. This is why I got into programming in the first place, before it became a career. It’s good to remember that.
And a small contribution to my language. ಅಕ್ಷರ-ಮಂಟಪ is probably useless in any practical sense. But it exists. A piece of infrastructure for Kannada in the digital world, however strange. Maybe someone will find it and smile.
V. What’s Next for ಅಕ್ಷರ-ಮಂಟಪ?
A few ideas, in no particular order of likelihood.
I might extract the bijection engine into a standalone Rust crate. Language-agnostic, well-documented, ready for anyone who wants to build a Library of Babel for their own script. Telugu, Tamil, Malayalam, Bengali, Gujarati. The Indic scripts alone could keep someone busy for years.
The grapheme cluster list could become useful for NLP work. 57,324 valid Kannada graphemes, enumerated and indexed. That’s not nothing. Synthetic data generation, testing harnesses for Kannada text processing, linguistic experiments. If anyone’s doing serious Kannada NLP and wants the data, reach out.
And the weighted distribution idea keeps nagging at me. A future post, maybe. “Making Infinity Sound Like Language.” The engineering would be interesting: you’d need to replace the uniform bijection with something that respects a probability distribution while maintaining invertibility. Probably some flavor of arithmetic coding. It’s been done for compression; adapting it for generation might be tractable.
But these are dreams for future selves. For now, the library stands as it is.
VI. ಅರಿವೇ ಗುರು
I keep returning to Basavanna’s line: ಅರಿವೇ ಗುರು. Awareness itself is the teacher.
The library taught me nothing I didn’t bring to it. Every page I found meaningful, I found because I searched for it. The library doesn’t generate insight so much as reflect what you bring to it. It’s a mirror the size of infinity, and most of what it shows you is noise.
But the act of building it taught me everything. The mathematics I had to learn, the linguistic structures I had to understand, the engineering problems I had to solve. The library was the excuse. The learning was the point.
Maybe that’s true of all projects. The artifact is secondary. The process is the thing.
VII. Thank You
To Jorge Luis Borges, for dreaming the library.
To Jonathan Basile, for making it real.
To Claude Shannon and Alan Turing, for teaching us that symbols can be computed.
To ಕನ್ನಡ, for being home.
To you, for reading this far.
VIII. Closing
I started this series with an aurora over Boston, particles dancing independently yet creating something whole. I end it with a library that contains everything and means nothing, unless you bring meaning to it.
ಒಂದು ಅಗಳಲ್ಲಿ ಒಂದು ಸಾಗರ
An ocean in a grain.
The library has been waiting for you. It always was.
ಅಕ್ಷರ-ಮಂಟಪ: https://sanathnu.github.io/Akshara-Mantapa/
GitHub: https://github.com/sanathnu/Akshara-Mantapa
ಧನ್ಯವಾದಗಳು