Base26
Software has always been a fusion of code and data. But for decades, we’ve been confined to two extremes.
In Software 1.0, we manually translated expert knowledge into programs, creating software artifacts that were 99% code and 1% data. In Software 2.0, a paradigm famously articulated by Andrej Karpathy, we automatically distilled vast datasets via training into neural network weights, creating software artifacts that were 1% code and 99% data.
The rise of Large Language Models introduces a third path, enabling us to program in the most intuitive interface of all: 26 letters of the English alphabet. What will the software of tomorrow look like? Will it look like Karpathy’s vision of a “LLM as a new operating system” or something else? Whatever it is, we do know this: The lines between programmer and prompter, and code and data, are blurring.
Base26 is built on a single, core belief: that the most profound technological progress will come not from “better code” or “bigger data”, but from the rigorous, creative exploration of the entire spectrum between them. Our work is a direct inquiry into the practical and theoretical implications of this new synthesis as we venture into this era (for full thesis , ). If you are a builder, researcher, or just plain curious, we invite you to connect ().