Wondering whether 10-12 years from now the internet becomes a hostile, corpo-inhibited medium, while the personal LLMs are what we take knowledge from. In reality some (of us) already do so, the only thing is that we don't really know what the LLMs were fed to begin with, but is assumed to be a reality-based information.
Your personal Search Engine is your personal model which you evolve to your needs. Safely including with your history, your chats, your family's memory. Internet then exists only as technical means to reach big corpo nets (heavily guarded against anyone extracting info from them) and as a ring of guerilla websites or fediverses, which are open by nature and resemble the old internet.
It is really to my opinion that this happens earlier than 2036.
I would not mind a LLM trained on say Wikipedia and its sources. That’s still a lot of bullshit, but orders of magnitude less than the open Web at large. The possibility to whitelist some websites would be useful. Training from the huge scientific articles databases would be ace.
What would be a strong selling point would be if the LLM is able to reliably cite where the information comes from. MS’s copilot currently attempts to do this, but often cites low-quality websites with questionable reliability.
Actually if AIs become good at reasoning, composition, logic better than humans, AI content will be sought-after and more trusted than human content for the sheer ability of AIs to better at those tasks than humans. Of course there will be need for human bubbles for "authentic" experiences but for most of technical content AI will be preferred for its ability to tailor it (for brevity, style, recipient's technical know-how, language, context).
How will AI discover new facts in the real world? What is the AI equivalent of boots-on-the-ground journalism? Of learning by doing?
A lot of reality happens in meatspace, and someone eventually needs to put that information into computers. AI can only rehash what's already there.
My question is who will do that work if you rob them of any satisfaction for doing it? A man running a widget review blogs could get reader mail, online friendships, advertising partnerships, a sense of making his mark. If everything a person creates just feeds an AI, what do they get in return?
Funsearch could be one way out for theoretical work. For robotics, we can have better simulated environments to learn, so no need for meatspace. Run experiments in simulated envs.
Your personal Search Engine is your personal model which you evolve to your needs. Safely including with your history, your chats, your family's memory. Internet then exists only as technical means to reach big corpo nets (heavily guarded against anyone extracting info from them) and as a ring of guerilla websites or fediverses, which are open by nature and resemble the old internet.
It is really to my opinion that this happens earlier than 2036.