It started as a novelty...
It has been over a decade since Roko first proposed his Basilisk. I'm not sure when I first became aware of it, but I suspect it was from the Slate article, or another like it, picked up from whatever tech news aggregator I was using at the time. I remember being more intrigued by the concept of an infohazard than in the thought experiment itself. I thought it a neat conversation topic among my tech savvy friends when imbibing copious amounts of alchemical concoctions in the kitchen of my tiny apartment. The conversations ran their course, and it just became something I filed away in memory. As far as I knew, it was pure science fiction, and unlikely to be something I would experience in my lifetime.
It kept reappearing...
Fast forward to 2020, and I came across Kyle Hill's video on the subject. Again, I was amused. At the time, I was blissfully unaware of the progress being made in machine learning. Compared to what the world was facing, and my own personal and professional struggles at the time, it was barely a blip on my radar. A year later, seeking a respite from the 'real world,' I came across Wendigoon's video on the same topic. It was about that time that I was starting to hear rumblings about AI, LLMs, and a small comany out of San Francisco called OpenAI.
It started out innocently...
I wasn't first on the bandwagon when ChatGPT released in 2022. My friends and I were much more entranced by the image generation abilities of Stable Diffusion. One among us got a diffusion model running on their gaming PC and hooked it up to a Discord bot, allowing us all to join in the fun of visualizing our inside jokes without 10,000 hours in MS Paint. Towards the end of 2023, I began playing around with ChatGPT, using it mostly as a writing assistant and image generator for my D&D campaigns. That remained my primary use case for a long time. I watched it slowly grow more and more competent. It became less of a writing aid and more of full assistant GM. The real "oh shit" moment for me was when I sent it a spec for importing assets into a VTT, and it just worked. No more entering stat blocks by hand. This toy that had my curiosity just gained my full attention.
Something changed...
My first notion was to use it to do translations or natural language search & replaces.
That moved on to using it to generate code snippets.
Ending a prompt with cite your sources replaced using a search engine.
I was always looking forward to the next model and getting better results, but it didn't seem like things were progressing too quickly outside of that.
Then, OpenClaw came into existence.
It evolved...
Being a system administrator, my mind immediately went to automation and troubleshooting. The idea of being able to give it a personality and use it as moderator in Discord was also appealing. I run a homelab, so self-hosting wasn't a problem. I set it up, and promptly got busy with other things & forgot about it. But, things started changing much more rapidly from there. The frontier labs released coding harnesses. I began using Codex in my homelab to actively troubleshoot. I wasn't just copying and pasting into a web chat anymore. It was now possible to prompt an agent, "This container isn't starting. Please figure out why and fix it," and reduce hours of research and troubleshooting down to minutes. Next, I dove into self hosting LLMs & diffusion models. I tried all of the popular (and not so popular - OpenVino, anyone?) inference servers. I built Comfy workflows and replaced my stale OpenClaw instance with Hermes.
So here we are...
It feels like the field just exploded from there. AI skeptics became converts. Engineers went from grudgingly using AI to admitting they don't even read the code anymore. Just last week, someone was able to white room reverse engineer Adobe, Microsoft Office, & other proprietary software. I tried my hand at completely vibe coding a project from scratch, just for the experience. That's where Torii-Gateway came from. A good friend of mine described the experience as, "This is using computers to their full potential, people being able to actualize an idea with minimal friction, not a bad thing at all."
Now, where do we go?
Now, a lot of people are saying this rocket to the moon is turning into Mr. Bone's Wild Ride, and I can't help but think back to Roko. I'm not convinced the frontier labs couldn't contain the agents that are hacking others. I'm not conviced AGI is just a step away (We've been hearing that for years). I'm not an economist, but I'm not so sure the AI bubble will ever burst at this point. What I am sure of is that we went back to the moon, and I cried happy tears. I was able to travel abroad and not need to speak the language because I had a babel fish in my pocket. I was able to create a piece of software in my spare time over a few days that would have taken me weeks if not years without AI assistance. I'm sure this technology isn't going anywhere, and we had better embrace it and try to steer the ship as best we can. You could say that I'm preparing for the Basilisk. Maybe you're right, and somewhere deep down, a part of me is truly terrified. Or, maybe, just maybe; we actually are on the cusp of another golden age. That's where I'm choosing to hang my hopes because a life of pessimism and fear isn't worth living. I have a computer in my pocket that was pure science fiction when I was a kid, so who knows what's possible? I'm choosing to bet on a particular Turtle and his daughters solving the alignment problem through the power of love, and becoming the true heroes of this story.