My Deep, Dark, Terrible Secret

This column is a reprint from Unwinnable Monthly #186. If you like what you see, grab the magazine for less than ten dollars, or subscribe and get all future magazines for half price.

———

Now this.

———

I have a deep, dark, terrible secret. Something I rarely share. Something only those closest to me know. But now, with a themed issue about AI, it’s time to share my secret with the world.

(Deep Breath.)

I know Sam Altman.

Well, maybe it’s better to say I knew Sam Altman. He was a year above me in high school and we played water polo together. He was the goalie, a position I always envied because I’m one of the laziest people you can imagine and swimming across pools was an ordeal, but my treading skills weren’t up to snuff for that position.

If my memory serves, he was nice enough. He never showed any signs of becoming one of our oligarchic overlords (although some other kids at our school did). So, because I knew him way back when, I learned fairly quickly that he dropped out of Stanford and bought a Ferrari after selling a startup. I also happened to be busy researching Reddit when Sam became the interim CEO, so I had been following his career a bit by the time he began getting coverage in the Times. I even emailed him after finishing my dissertation about maybe, possibly, perhaps, potentially . . . getting a UX research job at Reddit. After all, I was personally acquainted. I never received a reply.

All that to say, I was certainly paying attention when ChatGPT first arrived in the public arena. Suddenly, Sam Altman was a household name, as was the idea of generative AI. I dabbled a bit. In my day job, I’d been hearing about the rapid advancement in AI due to neural net and deep learning developments pioneered by folks like Yoshua Bengio and Yann LeCun. But I had also been learning a lot about the smaller projects – the generative work that doesn’t lead to environmentally devastating the world and enabling fascism.

While we refer to ChatGPT as artificial intelligence, it is functionally an advanced chat bot that is incredibly good at predicting what the next word should be. But the history to this model is long, dating back to at least to the 1960s when Joseph Weizenbaum developed ELIZA, an early large language model that, at least in some ways, successfully simulated intelligence.

Since the 1960s though, LLMs have been used for all sorts of different practices. One of the founding members of the Fluxus art movement, Jackson Mac Low, developed a model based on the collected works of Gertrude Stein and crafted a collection of poems with it. Mac Low was fundamentally interested in how chance affects linguistics and syntax. He was interested in recreating his “diastic” algorithm, which generated poems from two different inputs.

It sort of tracks, then, that some folks in hip hop are now starting to investigate AI as a tool for rhyming. Notably, in 2023 Lupe Fiasco partnered with the Google tool TextFX to create a new song. He sees AI as another tool to use to help create and given the long history of hip hop acting as eager adaptors of new technology, I don’t think he’s totally out of line there. Rappers and producers were into drum machines and sampling, confronting intellectual property laws and musical standards from the start, so moving into the space of generative AI doesn’t seem inherently out of line for the genre. Combined with the history of computational poetry, I think Lupe might be into something interesting here.

But, before you all jump down my throat, I’ll be clear that I’m not sure the juice is worth the squeeze. Cozying up to our tech overlords who are extracting the human labor and natural resources of the world in order to push an artistic practice into a new space is a mixed bag at best. By relying on Google (or ChatGPT or – shudder – Grok), Lupe or any other rapper is inherently tying their art to the tech in ways that can’t be undone. If he were, instead of relying on corporate tech, developing his own LLM from scratch and then building new songs from that, I don’t think there would be anybody who wouldn’t want to hear what that sounded like. If he followed Mac Low’s model – a small LLM, a limited LLM, one that didn’t rely on intellectual property theft of the world – I would be hard pressed to find a valid criticism.

And fundamentally, I think this is the way forward. Stop relying on ChatGPT or Gemini, big tech and our corporate oligarchs. Instead, let’s find bespoke, customized LLMs from independent creators that aren’t reliant on the benevolence of Sam Altman – a small large language model. Maybe then, I can stop seeing my high school water polo goalie on the front page all the time. That would be a nice change of pace.

———

Noah Springer is a writer and editor based in St. Louis. You can follow him on Bluesky @noahspringer.com.