Machine Learning (“mass training”?) didn’t exist 40 years ago. Cases like the tank one you described used a completely different technique that didn’t utilize RNNs or the like. Other than hardware capabilities, there’s been a big number of breakthroughs in the past 2-3 decades or so, from LSTMs to diffusion models and LLMs. It’s 100% not even close to what we did back in the 90s…