Back in 2020, a few months after Language Models are Few-Shot Learners was published, I was invited to a meeting with the then-Head of Product at OpenAI. We talked specifically about their newest model, GPT-3, and how I could use it at the startup I was working at. We had been selected to get early access to their Playground and API and had the ability to fine-tune GPT-3 for our use case. Our startup had one goal -- to provide educational access to the world via SMS (a method of information delivery that is much more widely accessible compared to Wifi or Cellular). We had developed a platform that let educators create a course that could be sent via SMS to their students in daily snippets. Up until this point, each course was handcrafted by a person. When I showed our CEO a demo of GPT-3 creating a course on neuroscience he teared up immediately.

That summer, spent in a "covid pod" in Oberlin, was particularly disorienting. A small college town that typically had most of its students return home or go to other cities for internships had even more of its year-rounders limited by the covid lockdown. Even fewer students were allowed to see each other, particularly indoors, because we had to limit the spread of the virus. This left me a lot of time to walk through the town and listen to podcasts or music or think through whatever problems I was having. This was back when the Lex Fridman podcast was simply called the Artificial Intelligence Podcast. I had joined Ben Goertzel's discord to discuss OpenCog, attended tea talks with Geoffrey Hinton, and worked through the Hands-On ML O'Reilly book. Between startup work and the loneliness of Oberlin I picked up a research project that would last 4 years on neural timing. My advisor and I would meet weekly and discuss ways to improve our model that would give neural networks the ability to have a sense of time without a central clock -- the same way humans can tell when a minute has roughly passed without looking at their watch. "Prompt Engineer" was a viable job title back then. We had no idea where this was going. AGI seemed more far-fetched and filtered by the rosy tint of in-the-future optimism.

Flash forward to 2022 and, as I passed through the entry hallway in UW-Madison's CS department, I was beckoned into a study room where one of my friends was TA'ing the Algorithms course. Another product from OpenAI had just released and both the TAs and students were using it. After asking it some basic questions, we probed its algorithm problem solving a bit. Then, we started to push it by asking for illegal information. It was like some sort of weird cyborg recreation of Rhythm 0 in Wisconsin.

I couldn't have predicted where the state of AI ended up today. Any kind of progress this rapid brings both inspiration and fear. The promise of 4 hour workdays through AI efficiency has been swapped with the expectation of twice the output. In software engineering specifically, new grads are struggling to find work as coding agents have largely replaced their ability at startups and Fortune 500s alike. For those with job experience, the time to sink or swim has already arrived.

I was hesitant to use AI to code. Frankly, I consider it lazy. I spent years studying Computer Science and honing my problem solving abilities. I was promised job security upon graduation. Now, I can order something to do my work for me, yell at it when it isnt correct, and send it to production without tapping into any of my frontal cortex. I shared my stance on AI with my coworkers: AI is like an amplifier. If you're lazy to start out, AI will make you more lazy. If you're motivated and curious, it'll help you achieve your potential. There is no other time in our history that we have had something like this.

I've used AI at work. And, like smoking a cigarette after a few beers, it feels great in the moment (when it works, which it's actively getting better at). But, in the same vein as taking a drag, I can sit at my desk and get hit by the reality that something that previously would take me 3-4 hours was just finished in 5-10 minutes. Am I suddenly bad at my job? A different kind of headrush emerges. It felt good at first and I have my own private calculus to justify it (I only have one a month; but it carries over month-to-month, cigarettes in Europe don't count, tennis somehow cancels it out, the social aspect of being outside and chatting over a cigarette _is_ healthy). The same calculus doesn't apply to using the coding agent, though. What have I gained? 3.5 hours back -- not if I just move on to the next thing. No stress that comes along with problem solving -- that stress is almost always good for me, and I've sacrificed it at the cost of having a deep understanding of what just went to prod. AI code that ships and breaks in prod is more of a pain to fix. The time to understand it thoroughly is shoved into the future until it's _actually_ critical to understand and fix whats going on. Not to mention every time I use it I wonder if it would be the last -- the AI learning my particular approach to problem solving via prompt engineering and retraining itself. I can wake up the next day with a sore throat and tar in my nose, declare that I'll never smoke again, and find myself outside at 2am the next week (remember the carry-over clause) doing it all again.

The same question comes up over and over again. Did I make the right choice?

Ten years ago, when I was applying to colleges, I was getting a lot of attention from art schools. I spent more time painting than doing anything else during the second half of high school. Oil painting was the epitome of creation. I could shut myself away from everyone else and bring something into the world that had never been there before. I could paint the same still life as someone, but the way I specifically delivered the paint to the canvas was special. I could communicate feelings visually that my adolescent brain and mouth couldnt. I did a summer intensive at SAIC. I was told that during the group show at the session's close the director of admissions was looking for me while I was off in other rooms. I followed up with him via email and included a photo of one of my pieces. He replied: "...my disappointment [of not meeting] instantly faded from seeing your painting from the exhibit again. So no small coincidence that every time a particular piece caught my attention to strike my fancy that your name was attached. Suffice to say, I cornered Magalie [my instructor] and told her how your art stuck out even among the others --- which is very impressive and no small feat in itself for a rising Junior". Magalie purchased one of my pieces a few weeks later. I never ended up applying for admission.

I wanted to do what I was good at and see how far I could take it. At the same time, I knew I was already good and could try new things. My parents talked to me about job security and warned me of a life serving coffee to pay for studio space. I wasn't great at math, but learned how to code with some confidence. I decided to go to a liberal arts school that would "let me paint /and/ take CS classes /and/ play piano" according to the interviewer (it wasnt). I told myself that I would always find time for my passion and, since college is expensive, I may as well study something that makes the most amount of money.

Will my job get replaced? Maybe. Do I have a chance to focus on the things that won't get replaced? Yes, and "chance" is more of a requirement to me than a suggestion. If a friend cancelled dinner plans with you last minute after you've already arrived at the restaurant, would you eat alone or call the whole thing off? I would almost certainly continue the meal. The plan has changed but I can appreciate it in a new light. Food tastes different when I'm alone, for better or for worse. I can take notes on it if I have my notebook. It's not rude for me to read while I wait. I can look up the wine I might have without worrying about being on my phone too much. I chose CS over painting because it felt like the safer bet back then. Now it's turning out to be the unstable one. It's like I've been handed back something I havent touched in ten years -- the chance to dive into my passions again. Should I mourn the potential loss of my and other's careers, or rejoice in the opportunity to establish my humanity again through the creative process?

For me, that means creating things that come from my own hand and have been filtered by my ears, tongue, and eyes. Inviting friends over for a meal. Practicing curation -- remembering the names of artists that I like or songs that have touched me. Practicing putting words to my opinions and deepening them past "I like/dislike this". Making a piece of jewelry or clothing for someone I love, knowing that when they wear it there is nothing else in the world that can also be that thing, and that no AI is trained on that specific pattern piece. It means protecting what I've created. Being careful about what I share and, letting my guard down low enough to be able to share it freely. It means remembering strangers' names and following up with them. If I spend hours of my day talking with something that doesnt have blood running through it, I owe it to the world and myself to counteract that inhumaneness with my creative output and connection with others.

---

Note: I picked up the habit of writing with em dashes in college. None of this was written by AI.