All #learning must be open-ended. The learning agent (the #observer) must have the #autonomy to set its own learning goals as well as plan and execute a #sequence of #exploration activities to achieve these goals.
One can never learn *all existing data* but rather refine their understanding of the data that is available to them. As true for human intelligence, you can either have "deep and narrow" specialized #AI agents or "average and broad" #AGI. You can't have both in the same entity. Time and #memory "limitation" are the main inspirations for #diversity and #cooperation between learning agents.
People should have figured it out by now that the #distribution of processing power, not the #centralization in gargantuan data and control centers is the right thing to do.
Stop working on LLMs (Large Language Models) and start working on PCAs (Personal Customizable Assistants).
From: https://arxiv.org/pdf/2311.00344.pdf