When ChatGPT came out in early 2023, its capabilities were shocking to me. Even though I was broadly aware of the latest in the field, and expected the capability to show up within the next decade or so, but I wasn't prepared for it to be there so suddenly. I spent the next few months thinking about it and expecting everything to change again, suddenly and quickly. The most immediate impact could be on Software Engineering as a discipline. No one should ever need to code again and AI should be able to do everything. And this capability seemed to be just around the corner for me.
As I dug deeper into the tech, I realised that it was probably not as straightforward. There were significant challenges. And some of them were fundamental. An AI which can code like a human, will require the building blocks of Sentience from the statistical underpinnings of LLMs. But it seemed to me that it should be possible and within reach. As I went down that rabbit hole, not everyone agreed that it is possible. Either sentience is more complex or the building blocks aren't feasible, yet. But it wasn't clear why, or why not. So I took out the toolbox, and along with a small set of 'believers' and started experimenting early this year.
Just yesterday we launched NonBioS.ai . NonBioS stands for Non-Biological Sentience, or Non Biologicals in short. We expect to path an alternate path to AGI (Artificial General Intelligence). We will build Sentience first.
Our first launch is nbs-1. nbs-1 is designed to solve a simple problem: Given a github code base, and a feature request, it will browse and understand the code base and give you the exact code to implement the feature. If it gets stuck, it will ask you for help. If you have a large enough code base, it will become proficient in it over time. It will learn to get better in implementing complex features autonomously. If the code looks good, you can ask it to check it in. You don't need to write code anymore. Just approve what nbs-1 writes.
This capability is fundamentally different from Code Co-Pilot's that exist today. Co-pilots help makes developers faster, but nbs-1 will replace your Entry Level engineers. For now, it will still require Senior Engineers to guide it. With nbs-1 we have taken the first stab at bridging the gap from LLMs to Sentience. We are rolling out the first constructs to solve for Agency, Attention and Adaptation. These constructs will eventually help us build Non Biological Sentience.
We expect nbs-code-1 to evolve over time and automate Senior Engineering Roles. The best/worst case (depending on how you look at it) is that this capability will decimate head count from the IT Industry. And it should be an year, or two away.
Related Posts
