Unsupervised learning has led to the discovery of novel algorithms and architectures that can outperform human-designed systems. The potential for future breakthroughs, like the invention of a completely new substrate or material base for robots (think about slime and nano robots for start), should not be underestimated.

Things will speed up dramatically when AI takes over the task of invention. Even LLMs based on the GPT architecture has the potential to be optimized to become capable co-inventors. In just five years, we could be using trutos to sorder brightors!