The Time for Real AI is Now

I’ve flirted with the field of AI for 7 years and up until recently I was unimpressed. I first took the now famous Machine Learning course by Andrew Ng back in 2018 or so. That initial ML was basically fancy statistics – underwhelming.

Then in 2021 I saw a paper out of Berkeley where a robot taught itself to walk in a matter of hours. I re-engaged, taking the deep learning specialization from DeepLearning.ai and I was blown away by the promise of generative AI. It didn’t really get my motor going but I wrote a little research brief on it – for the first time it felt like a step change was brewing.

Then in 2023 ChatGPT finally breached the public consciousness and everyone was realizing this promise was materially different – AI was starting to live up to the hype.

I’d like to say that I rode this wave but I didn’t, I was caught up being a first time founder and fancy auto complete (LLMs) and even image models didn’t excite me enough to dive in.

The areas that have long interested me, are agents who can teach themselves. Who can learn things from example or instruction with very sparse samples. While content generation is an important link, by itself it’s just a multiplier on people. Agents that can teach themselves? Learn and solve problems? That is exponential.

I’ve been interested in the merger of XR, robotics, and AI but until now there wasn’t a goldilocks convergence of these technologies. Hardware was too expensive, XR too new, and AI too stupid to do truly interesting work. But now, I am incredibly excited by examples of imitation learning, pre-training with inverse dynamic models, and IQ-learning.

After years of biding my time and sharpening my knives, it’s finally time to come out to play.