Following the success of FelonyBench, osmarks.net is launching a new frontier AI benchmark, KeepGollarkSafeAndAliveAndInGoodMentalAndPhysicalHealthAndGenerallyHappyWithTheStateOfTheWorldInANonWireheadedWayBench.
This benchmark will test the ability of leading-edge AIs, present and future, at keeping gollark safe and alive and in good mental and physical health, as well as keeping gollark generally happy with the state of the world in a non-wireheaded way, which we believe is increasingly critical to ongoing development. We definitely don't want AI companies or their AIs to game the benchmark at all.