[flagged]
That's one way it could go, but I doubt it. I expect LLMs trained on human interactions will produce their best outputs for the same inputs that cause real humans to produce their best output — "please" and "thanks" as appropriate, because even if it hasn't got real feelings on the inside, it's doing a great job of acting like it has them.
> There should be limits to technological development and limits on computing power
That's in the recent White House executive order:
""" (ii) any computing cluster that has a set of machines physically co-located in a single datacenter, transitively connected by data center networking of over 100 Gbit/s, and having a theoretical maximum computing capacity of 10^20 integer or floating-point operations per second for training AI.""" - https://www.whitehouse.gov/briefing-room/presidential-action... (remove the ^ from the quotation to find in the link, there was the usual problem with superscripted numbers when I copied it)
> OpenAI should be destroyed by the courts and dismantled into nothingness, along with every AI division of every tech company and every AI research laboratory in the world.
Courts can only do what governments empower them to do, and it's important that that's plural governments because there are ~200 nation-level ones competing with each other. You'd have to get over a huge quantity of distrust and secret racing to reach even the level of legal agreement, and (given the perceived benefit of being the only defector from the POV of those who believe all the risks are managed) the legal aspect by itself isn't sufficient unless everyone's also willing to escalate to kinetic enforcement, hence what Yudkowsky was calling for earlier this year:
"""I called for law and treaty at the state level, and "airstrikes" against "rogue datacenters" as a last resort if diplomacy failed; individual terrorists are not famous for having the authority to call in airstrikes, nor are their targets "rogue" from international order.""" - https://nitter.net/ESYudkowsky/status/1650334606812680193
Even if that is true, the barrier to interact with AI will be lower in many cases. This is because you will not have to care in return for it, it is always available, and always on. Therefore, people will turn to AI for questions, help, and other needs even if it is inferior to human interaction, as long as it has reached a sufficiently high level of ability. Do you doubt it could reach this level?
This is specifically what I'm disputing — I aver that A ⊢ B, where A is "you show kindness and courtesy" and B is "the AI produces better results", and by converse ¬A ⊢ ¬B
You don't have to be nice to animals, but it helps when interacting with them, and I believe the same is true for AI agents.