Ai is better than you

It’s an almost statistical certainty that whatever skill or attribute you, dear reader, may choose to consider, you’re simply, factually, objectively, not the best at. That’s in no way an insult, nor is it meant to put you down; it’s simply an observation.

This has always been the case¹ Note 1 × Disregarding, of course, prehistoric times, in which smaller populations somewhat upend my little statistical observation.. But it’s rarely ever enough to dissuade children from picking up sports, instruments, or an introductory textbook. Personally, I’m perfectly content knowing that most of the song lyrics I’ll write will pale in comparison to Bob Dylan or Gareth Lilliard’s works, and I continue banging my little electronic drums even though they don’t, and probably won’t ever, sound quite like Danny Carey’s when he plays his.

Although as of very very recently, I actually kind of can, in a way. I can - at least in theory, I haven’t put in the effort to fully verify this - ask chatGPT to give me lyrical musings in the style of Bob Dylan on the death of third spaces, and feed that into AI music generation software with a prompt to back it with prog metal drums. And after a minute or a few, it’ll churn out what would, done the old fashioned way, take at least several hours from highly talented musicians, a recording engineer, and a mixer/producer. All into a little demo track that can be heavily altered just as easily, or discarded only slightly more quickly than it was generated.

There’s a decent chance that the output would be fairly derivative, even if the words as a whole and the audio are all unique in the technical, dictionary sense. Still, one throwaway description I came across on a forum some time ago captured my own sentiments well - paraphrasing from memory: “derivative or not, the stuff it can put out would’ve been chart toppers a few decades ago.”² Note 2 × Perhaps the obvious retort is that they’d be chart toppers because they’re largely reproducing chart toppers they’ve been trained on. While there might be some truth to that, I feel as though that argument too easily handwaves the sheer difficulties of sound design, mixing, and producing that these models have clearly learned to do in the process.

In a drastically different context, a similar thing has happened with my profession and long time hobby, computer programming. I used to pride myself both on the speed at which I could churn out websites and apps, and my ability to find creative solutions to designs that went against the grain of what’s easily done in consumer apps. Now, while the debate is still to be settled whether today’s AI can produce code that’s as neatly structured as what I can, there’s no debate that with detailed enough prompting, AI can build any design I can in a fraction of the time, and with a fraction of the perceived effort (it’s been pointed out to me that I sigh and curse constantly while at my keyboard).

Which begs the question, why put in the effort at all? I came up in the style of snobby developers that took³ Note 3 × Intentionally past tense. immense pride in elegant, if not occasionally overly clever code. And while there are clear academic arguments that vouch for the importance of quality code, the codebases across most large, successful enterprise companies provide a deafening counterargument: the product (or going even further, the product’s commerciality), not the code, is what truly matters - cyclomatic complexity be damned.

And it’s not as though frontier models are incapable of producing elegant code; from my experience, they wholly are, but they require a bit of steering to do so rather than it being their default.

So, with all of this in mind, when asked to build another feature/page/website/app/tool/SDK, it feels absurd, in the literal sense, to do so “by hand”, the way I’ve done it more or less⁴ Note 4 × While the technical stack, programming languages, and frameworks may have changed somewhat over the years, the overall approach has remained the same. for going on a decade. The majority of yesteryear’s serious professional programming has been reduced to the equivalent of choosing to handwrite assembly over C for the sole purpose of handwriting it. Or, for any non-programmers that might’ve stumbled across my humble site, the equivalent of digging a hole by hand when an electric shovel (whatever that might look like) is handily available.

I’ve seen this sentiment repeatedly across the internet, though it comes in different forms. Most often, at least from converts like myself, it’s a variant of “I’ve never been so productive and yet so unfulfilled.” The task of building has, almost overnight, shifted to a role of almost purely supervision. For those whose delight comes primarily from seeing newly erected buildings, life has never been better; but for the passionate bricklayers who savored the journey, there’s very suddenly a massive void.

While I chose to use programming as my driving example, due in equal parts to it being an area I’m intimately familiar with and an area currently prioritized by frontier AI models, this is without a doubt a scene that’s played out before. When IBM’s Deep Blue beat Gary Kasparov, the pinnacle of achievement in Chess shifted from being the best chess player, to being the best human chess player; still very impressive, but surely a caveat big enough to take some wind out of the sails of exactly the type of individual to strive for #1 in the first place. Only now, the difference is just how widely and rapidly this phenomen is reoccuring across office work, math, science, and so forth.

Over time, I’ve come to realize that the disparity of savoring the journey versus the destination is the core of this knot I’ve been struggling with. This realization has made me more conscious of what the true goal and intention is behind my undertakings, my hobbies, my ventures. More precisely, it’s made me actively consider how much of my motivation lies in the journey, the destination, or a weird new third option that I’ll for now coin a “naturally-reached destination”, i.e. an achievement reached with no or deliberately limited assistance from these magical tools.

And it turns out that, for a healthy portion of the things I like to do, spending time handcrafting a suboptimal thingy is still more enjoyable than chatbotting my way to a more polished, technically impressive thingy. Not always, but enough that I still often enjoy being, and hopefully in the process becoming slightly less, naturally mediocre.