As I sat down for dinner after my workout, I found myself, as usual, thinking about Artificial Intelligence. I pulled up a YouTube video while eating — a talk by Ilya Sutskever from some 2020 conference. This got me thinking about how my perspective on AI has evolved over the years, particularly when it comes to Tesla’s AI developments. Back in 2021, Tesla AI Day kicked off. I remember watching the whole thing back then — I was still deep into the tech sector and a big Tesla fan. I was interested in AI and tech, but I was far from an expert. Hell, I’m not an expert now, but back then, I really had no clue how it all worked under the hood. During that first Tesla AI Day, Andrej Karpathy did a total rundown of their data labeling pipeline and self-driving architecture. At the time, I just thought all of this was cool and went on. I was young, a sophomore dealing with the start of the COVID pandemic and my military service. I didn’t grasp the full implications of what I was seeing. Fast forward to today, and I’ve been on a kick watching past Tesla AI Day videos. It’s funny how things come full circle. Now, when I watch those Tesla AI videos, it’s like a light bulb goes off. Everything starts to make sense. It’s hard to believe that was hosted in 2021 — the blood, sweat, and tears they’ve put in are really starting to show now. Looking back with my current understanding, I’m struck by how brilliantly Karpathy laid out the Tesla autopilot system. He broke down all the problems they needed to solve for driving — and there are a lot. You’ve got all kinds of edge cases in the real world, data labeling problems, and the need to specifically model the architecture to process lane changes, stop signs, and all that information effectively. Karpathy outlined all these architectural decisions needed to make the model learn. Now, with a bit more understanding of machine learning under my belt, I can see they’re really architecting the model to solve a specific problem — in this case, driving — and it’s far from trivial. It’s fascinating to see all those design choices: the data labeling and auto-labeling pipelines, how to perform all this computation on-device with very strict latency and power requirements. There are so many problems to solve, and back in 2021, Tesla was just starting to tackle these issues head-on. It’s incredible to think about how far they’ve come since then. Tesla’s stock has been skyrocketing recently. And I think it’s justified. Their Full Self-Driving (FSD) software is finally starting to do some meaningful work. When you watch the demo videos of people showing it off, it’s surprisingly human-like and natural. It’s mind-blowing to think this is the culmination of what the Tesla AI team has been working on for the past 5 years. I can’t help but have a few thoughts about all this. On one hand, I can’t really blame myself for not fully appreciating it back then. But on the other hand, I can’t forgive myself for not seeing the potential and going all-in on this stuff. I mean, just look at the Tesla stock now. If I had truly grasped what these guys were doing and invested, I’d be fucking rich. But here’s the thing. Remember at the beginning of this post how I mentioned that Ilya Sutskever talk from 2020? He was talking about GPT-2 and mentioned a demo called ‘talk to transformer’. Guess what? I had played with that exact demo on the web back in 2019! At the time, I thought, “Oh, that’s pretty cool,” had a tiny existential crisis, and then moved on. That’s the part I really can’t forgive myself for. If I had just had the courage to dig a little deeper, to grasp the full implications… I think I missed a real big shot right in front of my eyes. Now, I know I shouldn’t be too hard on myself. I didn’t know shit back then. I was just vaguely interested, not really invested in machine learning. But it still feels like a missed opportunity. So, what now? I need to get my shit together. Currently, I think I’m pretty capable of following and catching up with all the new stuff in this domain compared to my peers. But more importantly, I need to focus on catching those tiny sparks of something that could be really big. Something that might be small now but has the potential to explode. Or something that’s already big today but could become absolutely massive tomorrow. It’s time to pay attention and not miss the next big wave.