We built a machine that works like a synthesizer for a Large Language Model

What is it about?

Instead of the productivity-centered prompt-in, text-out interaction logic that dominates current LLM use, we propose a speculative interface that enables an iterative, hands-on writing process of assembling, plicing, looping, and editing. Users can turn knobs and press buttons to steer internal model parameters. Driving the calculations are adapted open-weights LLMs that run locally on the synthesizer.

Featured Image

Photo by Michel Didier Joomun on Unsplash

Photo by Michel Didier Joomun on Unsplash

Why is it important?

Rather than focusing on their ability to imitate a human on the other side of a chat interface, we believe there is merit in letting writers access and control the aspects of LLM language generation that are different from the human one.