Hi, I'm Jasper.

Working at the intersection of mathematics,
machine learning, and creative expression.

Go to https://dawntasy.com to visit my project website!
Explore My Work

About

Hi, I'm Jasper. I'm a 13 year old AI researcher, author and musical artist. My main project, Dawntasy, converges a trifecta of disciplines, ranging from philosophical research to engineering critical AI systems to deploying applications for everyday use. My latest book, Dawntasy, features a fantasy-dystopian world in which good and evil are reversed. Musically, I have been developing songs under the name "Dawntasy Melodic". Finally, my most powerful innovation involves my activity in the AI landscape, where I am actively researching deep learning paradigms and expanding the scope of what is possible. This involves training AI models, spreading knowledge and constructing applications such as AI agents and medical analysis.

My work includes...

Book

Published in January 2025, my book Dawntasy explores the connection between good and evil, and how fantasy and dystopian genres combine to reveal the shocking truths of the world that the characters are in.

The truth. Yaee craves it more than anything, yet it slips through his paws like a cruel illusion. In a fractured world ruled by the oppressive tyrant Time Smith, life and death mean nothing, but when Yaee's parents are murdered in cold blood, Yaee vows to avenge their deaths and seek out the mystery of his past. But memories don't add up. The world turns against him. The truth becomes a blatant lie. And the lie becomes the hidden truth. And that's when Yaee realises he knows nothing of the world and nothing of what he is about to face. In this cruel fantasy about a truth, the lie is inevitable, and that the horrifying truth is never meant to be found.

View Book

BiNLOP Paper

I am working on a research paper titled "BiNLOP", which investigates the properties of deep learning activation functions such as ReLU and SiLU. Specifically, BiNLOP addresses key stability concerns of the current paradigm of AI systems. By introducing a novel activation function with provable Bi-Lipschitz continuity while being highly efficient and expressive, BiNLOP demonstrates competitive empirical performance and stability against current SOTA.

LLM Series

In progress...