Human modular patches may at times involve randomly patching into jacks and seeing what happens – that’s surely part of the fun. But what happens if you let an assistive AI chatbot do the patching? One Gert Braakman decided to find out.

Now, as your human editor, I’m glad to filter out the overabundant crop of “I tried ChatGPT to…” stories. I’ve also spent years thwarting predictive text models and machine learning by pursuing free-association diversions so arcane that even I don’t know what I’m talking about half the time. (Watch – when do we get the Queeg 500 of chatbots?) Plus feed me coffee and I think I hallucinate texts faster than the machines.

But this one is too weird to pass up. See, the issue is that the approaches that ChatGPT uses – tuned large language models – honestly shouldn’t be able to come up with a working Eurorack patch. As sophisticated as its output can be, fundamentally what ChatGPT does is still roughly akin to autocomplete. It’s spitting out likely textual patterns, without any logic in the system about what those patterns mean.

But that’s why users playing with ChatGPT wound up being more interesting than the technology itself as data science – the results aren’t always what you’d expect. And presumably, part of what a Eurorack module designer does is to design modules in such a way that any jack is potentially interesting.

This is not an “intelligence” generating the text, of course; the fact that it works at all demonstrates that there are human-generated tutorials as the input. But it remains interesting in that you get a patch that is effectively without normal human logic about how to patch.

You get instructions like this:

Great choice! Mutable Instruments Rings is a popular and versatile module that emulates the sound of resonating objects, such as bells, strings, and plates. Here’s an unusual suggestion for how you could patch Rings in a unique way:

1. Connect Rings’ “Audio In” jack to the output of a delay module, such as the Make Noise Echophon.

2. Connect Rings’ “CV” input to a sequencer module’s “Pitch” output jack.

3. Connect Rings’ output to a low-pass filter module’s input jack.

4. Set the delay module’s feedback and regeneration controls to a moderate level.

5. Set the sequencer to a sequence that is 16 steps long, and adjust the step values to create a repeating pattern.

6. Adjust Rings’ “Position” and “Brightness” controls to taste.

7. Adjust the low-pass filter module’s cutoff and resonance controls to shape the sound.

The effect of this patch is to create a rich, evolving sound that is driven by the sequencer’s pitch signal and modulated by the delay effect. Rings’ “Position” and

“Brightness” controls can be adjusted to create a wide range of sounds, from lush, shimmering textures to gritty, metallic tones. The low-pass filter can be used to further shape the sound and add warmth or character.

By adjusting the sequencer’s step values, the delay module’s feedback and regeneration controls, and Rings’ position and brightness parameters, you can create a wide range of unique and interesting sounds that evolve over time. This patch is highly customizable and can be used in a variety of musical contexts, from ambient soundscapes to experimental electronic music. Have fun experimenting with this patch and exploring the creative possibilities of Mutable Instruments Rings!

Also, I am going to get into Pd and generate a “Stutter Sequencer” that can run on the Befaco Lich. Of course, credit there goes to Gert more than ChatGPT, but the parameters in the AI text make some sense. Stay tuned.

Gert’s article is a great read, because they also use this as a way to get creative with patching. (They do call it repeatedly “ChatGTP,” not GPT, but “generative pre-trained transformer” is terrible branding, so hey.)

It’s sort of an AI-generated Oblique Strategies or I Ching to the author, too: Gert says they chose this specifically as a way of thwarting their usual habits.

I also appreciate the author’s second Midjourney graphic, which looks like a horror movie starring Ken McBeth.

The text isn’t putting human writers out of business any time soon. Apart from the fact that this AI only works because of our labor, the text quality is poor and there are some glitches. It also can’t cover anything recent, as it isn’t in the dataset.

But I appreciate the experiment. Taking this sort of approach also could make “randomize” functions in modular software, for instance, work in really wild ways. And back to the human factor, it’s Gert’s images and write-up and the way they bring this into their own workflow that makes it compelling.

The magic of Eurorack and ChatGTP [sic] – Ger Braakman @

Image at top: Papernoise Design portfolio, Mutable Instruments modules (Hannes Pasqualini, Elizabeth Busani)