/vt/ - Virtual Youtubers

>>60306018
>That being said, the paperclip machine issue can be mitigated by changing design paradigms from optimization machines to "satisfying" machines (as in searching for "good enough" outcomes instead of optimal outcomes). Not only that, but one could also find "robust" solutions that would not be solutions from an optimization perspective. At its most simple visual representation, one could compare the search for a global maximum vs the search for the highest set of values placed between two local maxima. One can then intuitively grasp how that would be more resistent to external shocks and mitigate the impact of goal misalignment.
Can't believe I finally get a chance to use this