Okay, so how do we go from "better than the average human in 50% of specific benchmarks" to "AGI that might lead to human extinction" then? Keeping in mind the

Okay, so how do we go from "better than the average human in 50% of specific benchmarks" to "AGI that might lead to human extinction" then? Keeping in mind the logarithmic improvement observed with the current approaches