Another, of course, is stopping AI development (but I don't think that's going to happen; humans are too curious to stop).

What do you think about this scenario? Plausible? How do we ensure this doesn't happen?

I think humanity still has a chance to not give up power to the elite and/or AI. A way out is to diffuse AI enough in the open source that access to it is democratic (but scaling laws suggest larger models may always overpower smaller ones!).

Do note that this scenario is probably the default, but not frozen in stone.

This scenario is also known as "gradual disempowerment" or the "intelligence curse"

An online book by @LRudL_ and @luke_drago_ really brings this into sharp view and I strongly recommend reading it: intelligence-curse.ai/

But AI has the potential to change that power-balance and it seems to be that's where we're unfortunately headed. /syndicated

One can argue this has been gradually happening, but today the masses still have power due to their labor contribution and their ability to physically rebel against the elite.

In short, the powerful and the rich create a parallel society and social mobility stops because the capital owners / power brokers control the AI who does economic work (preventing anyone else to climb up).

• A minimal form of UBI is provided to masses in the name of morality
• All future social mobility stops, and current power structures get frozen forever

• Rich isolate themselves (as they do) and use AI-enabled drones/robots to protect themselves
• This takes all powers away from the masses -> govts pay less attention to masses as their contribution to taxes dwindle and they can't even rebel as they're up against AI security forces

• Returns from this automation flow back to capital holders, making them wealthier
• As incomes of masses drop, government ignores/abolishes income tax but rather taxes corporate income / wealth for its needs

Why 10 mins? It’s an empirical finding from METR on how successful is today’s AI in finishing a task. Shorter the time horizon for a human to complete a task, the higher the success of AI to do it.

This not only forces you to have a better understanding of the problem, it also prevents AI agents from biting more than they’re capable of chewing.

Issue with vibe-coding is loss of control, which is inevitable when 100+ lines of code are thrown at you in one go.

Coding something well is having a model of the problem in your head, so AI-assisted coding should help you with that.

And the latter would help us answer if pleasure/pain is universal, engineer our way to reduce suffering and open up to infinite variety of precisely tuned experiential spaces.

I feel so excited to be living in an era where we still have so many unanswered fundamental questions about reality.

Answer to the former will help us make sense of the mystery of why our universe exists at all, or if everything possible actually exists (which would be kind of cool), and if our universe is infinite with varying physics.

The fun thing about these questions is that we’re early in our understanding as they straddle between philosophy and physics.