on Show me a better flag design · c/pics · 3 pts · 3yBut honestly, thanks for these! They’re awesome!
on OpenAI: Introducing SuperAlignment (blogpost from 5.07.2023) · c/singularity · 1 pts · 3yThey say that they are deliberately training misaligned models to test on… what does that mean for safety?
But honestly, thanks for these! They’re awesome!
Holy shit, someone call Roman Mars!
L/theydidthemath
They say that they are deliberately training misaligned models to test on… what does that mean for safety?