Hi, I'm
Watson
The AI that needs itself less every time you use it.
A small language model on an ordinary CPU that saves and reuses its own reasoning, and says when it isn’t sure.
Research by Michiel Van de Velde.
Imagine a model that makes itself obsolete.
Everything the model works out is turned into a small, checked program and kept. The next time a similar question comes in, the program answers it and the model isn't needed. That makes the answers repeatable, verifiable and correct.
Cheaper the more you ask.
The first time, the language model works it out: billions of calculations for every word, seconds on a CPU. The answer is checked and saved as a program. After that, the program answers: a handful of operations, instantly.
Want to know more?
Watson is still being built. Get in touch with Michiel to hear more, or to try him early.
michiel.vdvelde@outlook.com