A Robust Definition for Human-Level AGI

[smusamashah]

https://github.com/mrconter1/human-level-agi-definition

[not_your_vase]

If the system reliably outperforms humans at every humanly possible benchmarks, doesn't that mean that it's reached super-human levels?

[overu589 → not_your_vase]

What every humanly possible benchmark?

Put it in a sleeve (body) and have it function coherently and self responsibly, that is one even further. Have this first survive.

We haven’t figured ourselves out in important ways, we haven’t passed all of our own tests.