Going Superhuman
(originally written Dec 2023)
Kamiwaza is a word that translates roughly to “divine skill” or “godly technique”. In a movie like Crouching Tiger, Hidden Dragon, the characters perform feats that seem superhuman; or take Ip Man, where the protagonist seems superhuman in efficacy.
This is the virtuoso we all aspire to be in different ways. It’s watching Peter Bence, who has the Guinness World Record for “most piano key hits in one minute”, bang out out a rendition of Africa (a la Toto) on a Bösendorfer creating all the percussion with a repeater and knocking on wood. It’s been years since I watched it on repeat and it still makes me giddy when I look at it. Or watching Jake Shimabukuro rock While My Guitar[Ukelele] Gently Weeps.
It can be different for everyone. Whether it’s martial arts, music, sports, or perhaps even good ol’ comic book heroes, we love to imagine ourselves going beyond.
I’d had some exposure to AI over the years. In 2020, I built a ~top-50 class supercomputer out of spot instances across 3 clouds in Northern Virginia, and used it briefly to crunch genomes using Parabricks, a GPU-powered adaptation of the Broad Institute’s GATK. I was reading rumors of the capabilities of a mysterious “GPT-3” in testing by OpenAI. I applied for access but didn’t get it.
By late 2022, I was hearing of people using AI more, building teams of data scientists to make predictions, and then I stumbled across the State of AI 2022 report. Reading about AI making the future happen across physical (tuning cold fusion plasma containment fields), ecology (optimizing an enzyme that could breakdown the PET plastic in a way that could industrialize true depolymerization-level recycling), and medicine (by proactively predicting the entire proteome via the OpenFold project).
I had an inkling to be part of that. It felt like magic.
And then… ChatGPT. November 30th, 2022. I’d been itching to try GPT3, so when this was release — and with an updated GPT, now v3.5 no less — I couldn’t wait. And yes, this story has been told to death. I was blown away.
At 8 years old, I first programmed a TRS-80. It came with a manual. I wrote a Wargames DoD computer emulator. Who would you like to nuke? North Korea, or the USSR? Then I wrote choose your own adventure stories for my little brother, and played with simple sprite graphics. But I’ve now read this story told by so many others in our grand community of computer science folks: there is magic that you instruct something, and it carries it out with diligence, precision, and grace. A tiny glimpse of what was to come.
When I discovered the Internet, my mind was blown. I’d dabbled as a kid with Compuserve and the Prodigy, but the Internet was next-level. Usenet already felt like a hive mind. It democratized information and industrialized communication and I felt that feeling again — superhuman.
It was the Internet that secured my life’s work for technology. I had planned on an MBA or a JD. But the Internet was magic — and I was to be a wizard.
That feeling was a long time coming again, but it returned on November 30, 2022. But much stronger now. And you know why?
Neural nets are magic. Not literally as you know — but as Arthur C Clarke said, “Any sufficiently advanced technology is indistinguishable from magic”. And so it is with neural networks. On a spectrum of “fully explainable” where you’d find something like a linear regression model, to “almost completely cryptic” you’d find ChatGPT and it’s CNN ilk on the far right.
Get mattwallace’s stories in your inbox
Join Medium for free to get updates from this writer.
You know something is reaching “sufficiently advanced” territory when the creators do not fully understand how it functions. Even speaking to some of the brilliant PhDs, they’ll describe it as, “more of an art”. We are like the Wright brothers, fumbling with wing designs, or Marconi, experimenting endlessly empirically to find the right way to make “wireless telegraphs” work. The Wright Brothers did not attack the problem with complex simulators and brilliant understandings of aerodynamics. Marconi was mocked by “real scientists” because he showed less understanding of the fundamental science, while being such a tireless empirical innovator that he was outpacing them.
It was with this in mind that I knew I wanted to do something in AI. And I have spent over a full year now immersing in it in a way I have never done in my life. At 5 in the morning, I wake up and want to code. I’ve done courses. I’ve written code. I read papers. I summarize papers with AI.
But I’m leaving the safety of ordinary work — leaving my job as a CTO, in a time of tech layoffs — because I’m convinced that this is the future. Like the superheroes of our modern myths, comic books, it’s like I woke up and found myself able to easily lift up a car… or float into the sky on wings of pure will. The span between thought and action has been compressed. The distance between intent and realization is smaller.
To Discord I go. /imagine prompt:a man, with the power of the entire universe in the palm of his hand
Magic. To ChatGPT:
Write me a single terse paragraph on why I should spend the rest of my life working on AI.
Dedicating your life to working on artificial intelligence (AI) presents a unique opportunity to shape the future. AI stands at the forefront of technological advancement, promising to revolutionize countless industries, from healthcare to education, and address some of the world’s most pressing challenges, like climate change and disease management. By working in AI, you can contribute to groundbreaking innovations, potentially improving the lives of millions and driving forward human knowledge and capability. This field not only offers intellectual stimulation and creative challenges but also the chance to be part of a transformative era in human history.
Well, I always wanted to be Superman. As a kid, I cried during Superman II when he lost his powers. Like, Dad-had-to-take-me-out-of-the-theater-because-my-sobbing-was-ruining-the-movie cried. We all need our heroes.
Within days, I’d done things in minutes or hours or days that would have taken me 10 times as long. I’d done things I barely knew how to ask for. It was clumsy — and sometimes it still is. But I learned all I needed to about this from comic books… when you get powers, you have to practice with them.
The thing is, we all have this power. It was a gift to ourselves. Standing on the shoulders of giants, with decades of math and computer science, with a huge portion of the sum of human knowledge now connected on the Internet, we created something marvelous. But I was a child of the Internet. It defined my life in many ways — through its connectedness to knowledge, to people, and how my career could not have existed without it. I remember how awful it was. How people called it a fad. How they mocked it as the tech collapse of 2000 tumbled the Nasdaq from an absurd high. How little that mattered to me, because I already knew it would affect everything. And even I didn’t imagine big enough.
I’m not making that mistake with AI. I’m not a naive optimist. But I also read in the tea leaves that this was a technology where the sudden leap took everyone by surprise. Now — with far less warning than the Internet — we are pouring untold billions into it. Tens of thousands, maybe hundreds of thousands, maybe millions of people are like me: practicing with their powers. It’s been one year since the ChatGPT moment. The advances are breakneck. They are relentless. The research paper releases are geometric. And yet the world is mostly untouched. Maybe we are closing the page on the “1995 of the Internet” now; many of us understand the world is never going to be the same. We’ve got the bug. But it will be years before the world is on the freight train of hype… and there will be a crash. And more years before the vision we have right now is realized. And then beyond that, there will be more.
I can only extrapolate so far… but I can extrapolate pretty far. Even if this technology only got better in sort of “fundamentally predictable” ways, like optimized software, smaller-nanometer hardware, etc, we haven’t come anywhere near scratching the surface of what we will do. But I’d hedge my bets toward more breakthroughs. Maybe a lot more. Change since GPT4 in March of 2023 has been steady but measurable. GPT5 didn’t suddenly pop out of the digital womb and cure cancer, diabetes, and world hunger in the first 10 minutes it was online — but there’s steady progress on all fronts to make AI, GenAI in particular, smarter, faster, better, cheaper.
So it’s in this vein that I’ve partnered up with my friend Luke Norris, who already had “Divine Skill” as an entrepreneur, to start Kamiwaza. “Divine Skill”. “Godly Technique”. Because that’s the feeling we’ve had. Suddenly our balance and grace is so flawless that we glide along the rooftops, and our katas move the wind. I want everyone to feel this. This divine skill. This power to impact the world, amplified.
I am writing this on the eve of that journey. Whatever happens, I feel like a man possessed. I want to take you on a journey; my own journey of discovery. A catechism of experiments and of building. I am a technologist, and this is my testament. Along the way there will be building. There will be experiments. There will be failure — and success. I don’t know where the journey ends, but I feel like this was one I was born to take.
