hi, i'm Pranav. I am the founder of Meteor, a Y Combinator alum and a CS dropout from the University of Washington, Seattle. I love building cool stuff. Within a week of launching my first project at UW, nearly ten percent of the undergrad student body were active users. It was immediately taken down, but that sparked my interest to build and I have since continued to build things that excite me. Outside of work, I enjoy chess, philosophy, AI research and recently taken an interest in mechanistic interpretability.
In this blog I plan to include (1) research demos I build, (2) condensed explanations/ opinions on new AI things that I read about, and (3) philosophy and other miscellaneous takes. I will try to update this website when I can. It is not a comprehensive account of my work or my interests. If you want to know more, please reach out to me.
In making this website, I also made the (ironic) decision to entirely avoid the use of AI. I am a strong believer that AI writing dilutes signal -- I find a lot of value in the rough, unpolished, semi-correct analogies and intuitions humans use to communicate and express the ideas in our brains. All words are my own.
Currently I am most interested in human taste. What makes certain arguments seem convincing? Or certain animations "look right"? What makes some texts feel insightful while others slop? I am also interested in Coconut (increasing latent reasoning in LLMs), continual learning, building a computer through LLM weights(???). Also, I predict that withing 5 years JEPA is going to be a breakthrough.