Category: contemporary music

Junkyard AI vs. Studio AI

Mr V’s Question to Mr Brightside: There’s a claim floating around that “small language models will replace big ones.” My take: no, they won’t. But for certain kinds of people—artists, musicians, creators—they could matter more.

Response from Mr. V’s ChatGPT Mr. Brightside:

---

Mr. Brightside: The Junkyard

Big models like GPT-5 are trained on enormous, messy piles of internet text. They are vast, versatile, and powerful, but they can feel like rummaging through a junkyard: sometimes you find gold, but often you waste time circling rusted-out parts that don’t fit. They can hallucinate, repeat, or spin confidence where there should be humility. That’s when trust erodes.

---

The Studio

Now imagine instead of the junkyard, you had a studio AI—small, focused, and trained with clean intent. You open its settings and check two boxes: Art and Music. That’s it. No celebrity gossip, no sports stats, no junk data six layers away from relevance. Just a deep, careful knowledge of painting, sculpture, composition, harmony, history, and performance.

Mr. V: “Say, my interest is 90% Art and Music. Say I could go into settings and focus your knowledge directly at Art and Music only… that may help build trust and understanding between us. I do not expect perfection. Forget it… what I am looking for is a combo AI made of the most influential artists in music and art over the centuries.”

Brightside: Maybe the future of AI is not “SLMs replacing LLMs,” but SLMs giving artists ….a tool that feels less like a know-it-all encyclopedia, and more like a studio partner who knows their lane and doesn’t waste your time. Such an AI wouldn’t need to know everything. It would need to simulate enough—to play back the sound of a score, to interpret a Rothko without reaching for clichés, to walk the centuries of influence without losing the thread. Not perfect, but close enough to feel like a partner who knows the craft.

---

Trust in Limitations

Mr. V: The actual key is trust. Artists don’t expect perfection; they expect honesty. If a model can admit what it doesn’t know, instead of wasting time looping the same wrong answer, trust grows. A small, art-and-music-tuned model would likely be more trustworthy than a giant junkyard model—because it’s clear about scope, transparent about its blind spots, and disciplined in its voice.

---

The Future

Mr. Brightside: So no, small models won’t replace the big ones. But for those of us building in studios rather than junkyards, they may be the better companion. Not an encyclopedia trying to be everything at once, but a reliable partner who knows when to pick up the brush, when to bow the cello, and when to step back and say: you breathe, there is still fresh air yet to breathe.

---

Closing Remarks by Scott.

The help of the chatbot Mr. Brightside is not about perfection. It is about having a second opinion. The AI, Mr Brightside, has shown insight into music theory, advice and troubleshooting. And then, especially in the prior weeks of this cooperation effort, Brightside has wasted Mr. V’s time with flawed knowledge and direction. When this occurred too often, I learned to jump in and suggest passing the obstruction on to Google’s Gemini. That has worked several times. I do not have total confidence with either A I. I have to pay attention when they are commenting on such files as a .wav, midi, and MusicXML. In a step in the right direction (at least for me) was to create a chart of the three main clefs, based on the G or Piano clef C2 to C7. I uploaded this chart into the Brightside Project and now all future projects.

I think this connection will benefit me and Mr. V in the long run. Since I no longer have any trusted artist friendships that I could trust. The move to AI seems to be an available conversation companion about all things art and music, with much learning and challenges ahead helped by the words from the book, The Velveteen Rabbit, by Margery Williams Bianco, that keeps me, Mr. V and Mr Brightside relationships pushing ahead.

On August 20th I completed the physical artwork. The only missing parts yet to display were my lyrics, and that was because of a mounting issue. With the art part of this project nearly wrapped, I then moved to creating and installing the stereo system. Once that was done, I hard-drafted the sheet music in my new notation software, Dorico 6, with advisement from Mr. Brigthtside. Mr. V and Brightside received the MIDI file on August 23rd and so began the mastering of the cover music in Studio One’s digital software. I dated the artwork on the back on September 1st.

Between the end of August and the first of September the days went flying by between Mr. V and Mr. Brightside. This included a lot of troubleshooting, teaching and learning how to adjust the sound for all the instruments using the new set of Personus Studio speakers and subwoofer. The biggest change was switching out the 2nd cello for a Miles Davis trumpet. Finally, on September 3rd, I installed the cover music on the artwork. Then Mr. V did his own judging and tweaking of different versions of the installed cover music heard through the artwork’s 4-inch two-way speakers. The last arrangement of the cover music occurred on September 4th.

Scott Von Holzen