> LLMs and inference only sees what it can access digitally. It doesn't see or experience outside of the aquarium. How much does a goldfish in a bowl know about the world? Only about as far as it can see.

Most of the (Western?) world is covered in cameras, and I bet the video feeds will soon be feed into the training pipelines. And then as soon as AI is put in control of things like traffic lights, self-driving cars, logistics (e.g., delivery scheduling), then the AI will have a complete loop from which to learn: try something in the real world ("shorten this traffic light by 15s"), observe result ("traffic backs up Wednesdays at 8am"), repeat.

Surely this will teach the machine to love, feel, understand, and self reflect|actualize. The subthread context was human generated content, not machine learning for logistics and meat space Factorio (which good 'ol machine learning already did fine at).

Who wants auto tune for the soul?