What do you currently consider the best giving multiplier for X-risk/AI safety? (highest & somewhat robustly supported by evidence)
What do you currently consider the best giving multiplier for X-risk/AI safety? (highest & somewhat robustly supported by evidence)