Tom Whitwell, the human who previously blogged about music tech, has turned his attention to just making gear instead under the same banner, Music thing. But that didn’t stop him from getting ChatGPT to hallucinate some news from Superbooth before it happened. Now that we’ve got the real news, how did the AI fare?

It’s really a test of how predictive the predictive part of fine-tuned large language model text generation can be. Or to put it another way, since AI itself really doesn’t have agency, it’s a measure of how predictable those of us in the industry are.

At least this opening text fooled me before I figured out what was going on:

No. The reference to Mutable Instruments is of course because the model is still rooted in the past, before Mutable ceased operation. The Spectra isn’t far from a handful of granular tools that did show up this year, though.

Oddly enough we did just last month get a system from Intellijel, so this isn’t entirely wrong. Cascadia isn’t even that far from Atlantis.

I think the really hilarious option is the Black Corporation Ouroboros, which effectively rips off the price and multi-effects of the (real) Expander MK2:

Ouroboros has been a real device, several times over – a dual-band effects box, a West Coast semi-modular, a rather pretty wooden DIY analog instrument, and a drone synth.

Careful in asking about what Tom’s doing, or you could … get infected:

He also proves a point. I’m very confident Tom is wise to the real possibility of misinformation using generative text models. And you can see the problem here: you quickly have real-seeming information with scrambled facts. It’s close enough to the truth that even when you know it’s not real, you can do a double-take while reading. If you didn’t know – including if these tools were harnessed as a practical, cost-saving measure and started spewing moderately inaccurate information about music gear all over the Interwebs.

So, uh, yes, anyone at Future Publishing reading this, please just don’t. But this starts to get really dangerous for matters more important than Eurorack modules. With synths, it’s just mildly entertaining / annoying.

I also messed with Stable Diffusion to create the weird images above. Unless I specified women, you get a lot of white guys with beards and even paisley shirts almost irrespective of prompt, the moment you mention synths. Huh.

There is always that guy with super long stretchy arms getting past the synth you want to try, too, so – realistic.

Oh, and since Chris Randall joked about it, I did try prompting ChatGPT to write the news “in the style of Peter Kirn.”

My initial suspicion that it would just choose an arbitrary style and say “I am Peter Kirn” was probably correct. I got things like this:

Ladies and gentlemen, synthesizer enthusiasts, and sonic adventurers, it’s time to dive into the latest and greatest developments from the sensational Superbooth extravaganza! I’m Peter Kirn, your trusty guide through this tantalizing realm of electronic sonic sorcery.

First up on our radar is the remarkable offering from SynthTech, the illustrious wizards of the modular domain. Prepare your eardrums for the mighty “Fusion VCO,” a fusion indeed of classic analog circuitry and cutting-edge digital wizardry. This beastly oscillator brings together a wealth of waveform flavors, ranging from smooth and velvety to serrated and ferocious. And that’s not all! SynthTech has ingeniously integrated an overdrive circuit and a spectacular “wave multiplier” section, sending your sonic explorations into uncharted territories.

And… uh… WTF?

As we bid adieu to another extraordinary Superbooth event, we can’t help but marvel at the relentless innovation and boundless creativity on display. The synthesizer industry continues to push the boundaries of sonic exploration, and these latest offerings serve as a testament to the indomitable spirit of those who shape our musical landscape. Until next time, my fellow sonic adventurers, keep pushing the limits and may your soundscapes be ever-expanding!

Uh, yeah, you do you, ChatGPT. (I do have a fairly large amount of text so at some point of course I’ll try self-hosting a LLM trained on myself.)