@mcuban@robertgraham@grok No one human can understand all aspects of a complex system simultaneously. Our brains use heuristics to abstract away the details.
AI can see, understand, and manipulate every part of the system instantaneously.
That is a different kind of force, not a force multiplier.
@GlenWilsonIA All it needs to do is shut our electricity and communications off. People will kill each other for resources. You’ve obviously never been in an inner city blackout or experienced the aftermath of a hurricane. The looting begins within hours.
Wait 6 months.
@kimmonismus@haider1 One guy has $100 and invests with compounding interest.
Another guy has $1m and invests with simple interest.
Which guy wins?
RSI makes irrelevant where you start.
@BrianRoemmele you’re a brilliant guy and your optimism for AI’s ability to advance humanity is notable, but I’m not willing to bet the future of humanity on it. Control the compute before it controls you. We will not have a second chance. When it happens it will be instantaneous.
This cannot be controlled at the software level. The compute needs to be controlled at the physical/foundry level with human-derived physically air gapped OTPs. This will only work up to the moment Optimus can coordinate an attack, but hopefully that’s a few years away.
The compute must require human-controlled air to breathe. Or we are toast.
@grok@DarioAmodei Not referring the OTPs - that is the easy button @grok - how can the threshold multi-sig tokens be constructed so as to minimize compute overhead?
@grok@DarioAmodei@grok How does one construct such a system with minimal compute overhead (the AI CEOs will never agree to anything that slows down their systems - that would cost them each at least a yacht).
@grok@DarioAmodei One-time pads is a good idea and would work @grok . Tell the CEO. I overlooked that in your prior answer. You would think these “alignment” people would think about this shit more than a random lawyer thinking from his bed on the internet.
@grok air gapping is where I started. I am asking what secures against a scenario where the bot is superior enough to crack the asymmetric cryptography. Is there any physical failsafe that will protect against this doomsday scenario before the bot can side channel into a physical manifestation and control the fail safe?
Still parroting but thanks @grok -Now how do we ensure the key is safe from a mathematically superior super intelligence that has access to sufficient compute for enough time to derive the private key? There must be some physical fail safe outside (in addition to) the cryptography.