(It's Not) AI Slop

A possible strategy for responding to AI bigots

By Tom Cully

So, 0.5.0 of the corvus-radioberry project was published tonight, and, almost predictably, the first comment on Reddit reads thus:

If you don't like AI slop (like me), you can use the original driver instead:

https://github.com/pa3gsb/Radioberry-2.x

It's interesting. I could go on at length about how every design decision was made by me, how all the code was read and checked by me, and how there's no technology or technique in here that I don't fully understand - well, that's mostly true. Let's just say I've learnt a lot about Verilog in the last few days. But it's more this: if the commenter had bothered their arse to actually install and test the driver, they might have noticed a few things, depending on whether their platform was an RPi4 or RPi5 - things that are impossible with the original code.

For a start, it works on an unmodified, non-custom RPiOS (Linux 6.18 Trixie) RPi5. This is because the original driver depends on a patched version of rp1-pio to handle the PIO vs GPIO data bus from the FPGA to the Pi. The new gateware rewrites the FPGA/RPi interface, moving from the Hermes-Lite-2 radio core to the driver over I2S. In fact, the hardest part was making sure that ALSA, the Linux distro's sound system, didn't see the I2S interface and decide that the RadioBerry was a sound card.

They might have noticed that the receiver-count lock-up in the original gateware is fixed, because the new images never clear a FIFO, and that control now uses a framed protocol with CRC8, with sequence checking on the transmit stream. They might also have noticed, prominently, that you can run up to six receivers at 192 ksps on a Pi 5 (four on a Pi 4), over your choice of Protocol 1 or 2, with piHPSDR at no more than 13% CPU at 192 ksps. They'd see the original software running a single receiver at 30 to 50% per core - yes, all four of them - at the same rate.

They might even have seen the discussion on the RadioBerry Google Group, where Johan, PA3GSB, the original designer of the hardware and author of the original driver, described my previous 0.4.1 implementation as "Very nice work. Very pragmatic implementation making a SPI device with PIO implementation." They would also have seen my ongoing engagement with Johan and other users. Even if they'd only looked at the Github Repository they might have noticed the 99% code coverage, the passing CI, the CodeQL rating.

They might, in short, have not dismissed my use of AI out of hand. I understand that the world is, in actual fact, drowning in AI Slop. I agree that caution must be exercised, especially when a project or piece of software is new. They could simply have tried it and seen that my 30 years of software engineering experience enabled me to drive AI in such a way as to produce something that was indeed developed with AI, but is far from vibe-coded garbage. The usage of AI in development doesn't automatically render the output AI Slop. I'd agree that, to use AI effectively, you really need to know what you're doing first – but in 2026, being a wholesale AI bigot is simply asinine. You can throw your clogs in those darn newfangled machines all you want, but it's here to stay. The best we can do is make it work well, and if you think otherwise, you're simply fighting the tide.

I could have defended myself, my experience, my methodology and my project at length in the comments. Instead, given that I have professional quality, tested - and above all, working software which happens to have been built with AI - I chose much the same words to deal with an AI bigot as a parent uses when their child suddenly becomes a picky eater:

"You don't have to use it."

It's no surprise to anyone that we live in this world of the late 2020s where we have absolutely no shortage of people with strong opinions on subjects that they – and I feel I'm being quite charitable here - know absolutely fuck all about. At least, as I was reminded this morning, negative engagement is still engagement. Maybe someone else with an actual problem, as opposed to an axe to grind, will find corvus-radioberry, enjoy their AI-enabled, trouble-free RadioBerry, and appreciate the cool CPU while they're at it.