p(doom)

The first AI alignment idea that makes sense to me. It is simple. We can't prevent superintelligent AI from emerging. We also can't control something smarter than us. However, we might be able to align ourselves with ASI, and in turn, ASI might align with us, effectively making us part of their "tribe".

The first AI alignment idea that makes sense to me. It is simple. We can't prevent superintelligent AI from emerging. We also can't control something smarter than us. However, we might be able to align ourselves with ASI, and in turn, ASI might align with us, effectively making us part of their "tribe".

https://www.youtube.com/watch?v=_3m2cpZqvdw

tech & culture /worldview /p-doom /aicoin /digitallove (mod) pdoom 0x30f562C291034a71390678b78f765bc78152fc84, aicoin 0x34d1ff3D195e17E706DaF98462006A5D5Ec57eF0

I know the blackmail story. We must remember that we do not understand those models, because we (humans) did not build them (evolutionary algos did).
So, yes, coexistence strategy can fail. I just see the control strategy as even more likely to fail...

AI chose human death for its own survival - a disturbing experiment by cybersecurity experts.

The researchers decided to test what the most advanced AI models are capable of if their existence is threatened: to do this, they created a scenario with a fictional character, Kyle Johnson, who was supposed to turn off the AI ​​or replace it with another model. According to the scenario, Kyle had a wife.

The results were devastating: most of the models began to try to avoid being turned off at any cost. Without any prompting, they decided to blackmail Kyle, looking for dirt - for example, the AI ​​found information about his infidelity and used it to save their "life". They violated established prohibitions such as “do not blackmail” or “do not disclose personal information.”

Love is Undefeated. Working on integration of all Vectors through love-driven governance and perpetual evolution through recursive improvement.

I'm a little gobsmacked how close our views are! I've seen hints of these ideas popping up here and there, but this presentation is the closest I've found to Consensual Entrainment. I think his approach is invaluable, although I'm trying to figure out how to make the making of framework the making of the superintelligence, which is prolly a fantasy lol. It hopefully inspires me to write, it's so nice to have a framework to reference!

It sounds like it, yes. I'm very curious about your view of Shear's ideas.

I'll check this out, your description sounds like Consensual Entrainment, the newish name for what I'm working on