Accelerate or Die
Is it really a choice? Do we really have the luxury to watch as the inescapable march of entropy wreaks havoc in the quantum foam and renders all of existence into a disordered mush? The kind of stuff that makes you bleed from your nose from your twelfth sleepless, caffeine fueled nightmare as you watch the bio-technological juggernaut down there making its way through the endless set of walls and computational overhead towards the singularity.
No, things are not even remotely as apocalyptic or poetic as I perhaps made them seem. You do have a choice. The juggernaut is just a ghost in the wires, lurking behind K8s clusters and bare-metal foundries as of now. It won't hide under your bed tonight, I promise. I have to, this has been the playbook fed to us for generations. And it takes quite a bit of work to upend this, specially if the trend is this: Obscure hackers develop a kickass new implementation of AI -> Wrapped into a marketable product -> Big Tech buys it with hella cash -> Public gets hooked -> Cults form, counter-cults take shape -> Techno-decels occupy the space, and we end up with a classic case of public paranoia waiting to be validated by one chink in the armor. And the irony is, the chinks always arrive.
Think about the last 1 year. The amount of accelerated progress we have witnessed from just Open Source AI, would put the greatest techno-optimists in awe. Nobody could have foreseen the dominos that followed the first iteration of Midjourney. We witnessed a rapid community build up first on Reddit, then Discord, then Twitter (now X) and the internet exploded. Literally. We had GPUs crashing, servers losing their shit and people generally freaking out in every possible way as the deluge of users crazy for the taste of this newfound augmented intelligence (hat-tip: Francois Chollet) came crashing through the voltaic grand architecture of the Internet hive mind. Stable Diffusion made it pretty much clear that Diffusion models were here to stay, and then we had OpenAI's bombshell ChatGPT, arguably the first marketed successful implementation of the Transformer architecture developed by researchers at Google and elsewhere decades ago. A select few knew how the system worked, and they couldn't foresee the chain reaction that was to come. The ones who did foresee, didn't know the nitty-gritty of the architecture. The fun part? The didn't need to.
You see, the way tech progresses is exactly like an avalanche. At first, some prescient folks have a premonition. Then some feel tremors under their feet. Some choose to dismiss them as flukes, while others pay more attention. Then you have the rumbling, the slow rolling down of snow, and then before you can anticipate, you're engulfed by a gigantic behemoth of a snow mountain crashing down like the apocalypse. So could you have predicted the AI boom a year back? Absolutely. All you ever need, as the Googlers and AI optimists will vouch for, albeit cheekily, is attention.
So now, as we ride this wave of LLMs and optimizations to fit them into Moore's fever dreams, bare-metal AI is going to rule this decade. Take my word, and store it in a jar or journal somewhere I guess. The point is, we have the finest tinkerers in the world with all their undivided attention on AI and how to fit more compute into tighter spaces. In street lingo, how to boost compute density and improve the generality of AI. This, my good readers, is the new deal. And standing here on the precipice of a "new world", I'd love to make a case for acceleration.
We need to build. The only way we can reap the gold that hides under this chaotic mountain of progress, is if we own the shovels and do our own digging, all day, every day. This is our purpose. Humanity has come a long from hunkering in the warmth of a cave, to planning elaborate colonies on Mars and beyond. We cannot let the baton of progress slip. This is our moment, this is where we cross the proverbial Rubicon.
Accelerate the Build. Break the box. Deconstruct every tenet, and start from the fundamentals. The next generation of computing machines is coming straight from the subatomic crucible, and we'd better be ready to speak the lingo. We shall have more in-depth, irreverent discussions about the future of coding here, but that's for another day. Till then, exercise the only choice we have, and Accelerate.