Contemplating slop; out there and in here
or Between The Rhubarb And The Stars
02 Dec 2025 - atrodo - Song: Between The Pavement And The Stars by Five Iron Frenzy
This year has been filled with a good amount of effort to produce a little bit of progress. But as the world of LLMs (colloquially marketed as AI) has started to accelerate with capacities and new aspects, I am starting to find the things inside me that that used to motivate me start to drift away.
So to start with, like I mentioned in the last post I had been working on rebuilding robot brains to think in 8 note scales instead of 12 note scales. I can happily say that most of that work is complete. This came with a good number of improvements that made things easier in general. The most of important of which was coming up with a reliable way to encode and decode scale numbers in a reversible manner. This drastically improved the process of comparison of notes by making everything that talked about notes and note differences use the same process.
The core of this effort was two functions, cmp_notes and trn_notes, to
compare two notes and produce a difference, and to take a difference number and
translate it into a note. As part of process, I wrote extensive comment about
how that behavior would operate. This detailed how, both functions would behave
with certain inputs, even going so far as to give concrete examples on the
behavior. After it was done and proven to work, I gave those extensive notes to
an AI, asking it to produce code following the instructions. Maybe you’re asking
which LLM, and to you I would say that doesn’t matter for the story.
The LLM produced two functions much longer than mine, so I gave it my translate function and asked it to compare. It said, and I quote, “Your implementation is much cleaner and more mathematically elegant”, “Your implementation is significantly better” and “This creates a …, which is brilliant.” So I asked it the obvious next question: “If my method is so brilliant, why didn’t you use it?”. It responded with “You raise an excellent point. I didn’t use your approach because I simply didn’t think of it - I got caught up in trying to directly interpret your prose specifications without stepping back to find the underlying mathematical pattern.” Gee, that makes me feel better.
But that doesn’t stop hundreds of millions people from using these LLMs to produce slop. As the world of LLMs explodes, there is becoming an increasing amount of “slop” being generated.
Let’s start by talking about slop. I would define slop as low effort creations utilizing machines, currently with LLMs, to generate as much as possible in an attempt to capitalize on the technology. Some of this slop has started to appear on youtube and spotify in the form of machine generate music. I’ve seen mixed responses from people; some enjoy it because it produces for them a style of music that isn’t being produced like rock from their formative years. Others seem to find the music soulless and a con, raking in money from these platforms by only typing in some words and uploading. In the end its all the same, slop music being generated by computers.
If we use that definition of slip, that is exactly what teaching robots to drum is trying to do. The main goals of this project were, in essence, to generate good enough music to entertain people enough to listen. I could say that the robots route produces music that is more interesting than the rest of the slop because it was hand curated by me working diligently to create a system from the ground up. I could also ask, does that matter? Is this project just adding more machine generated slop to the system?
Obviously there is value, to me, to generate the code. But without a goal that I believe in, does that means am I at the end of the road? Have the robots now provided me enough value in the experience of creating it that there is no longer meaning in working on this? Has teaching robots run its course? And is there value outside of the value that I create for myself by creating this? I don’t have an answer for that, and I continue to ask myself that now.
I started writing this blog almost size months ago. Its been a struggle to write; more than usual. I started this blog to share the progress I have as a form of self documentation and to share insights I have obtained. Looking at the analytics of the page, it appears that the blog has not reached any kind of audience. Technically, the blog has more views than the videos the robots have produced, although the data that statement is made from is suspect. Looking at analytics for the page, so much of the traffic is on windows 7 and an extremely specific screen size, which makes me think most of the traffic of this blog is from bots scrapping my content. That seems ironic or poetic to me.
And all of that is really saddening, and I’m very much unsure where this leaves Teaching Robots to Drum. This project has always been about learning and producing something. But what point is there when nothing I can produce can be appreciated by any other living being. What point is there when anyone can produce a better version of the robots just by asking some website to do it for them. Why should any of us proceed creating things for anyone else when some giant tech company can let anyone do it.