I know what I would do to build one. But I don't think I should talk about that. The questions we should ask now (my contribution to AI safety fashion), if how do we upgrade our baby-AGIs. If the AGI of the future/present is a dominant entity, and it is judging our behavior, maybe we should make sure not to kill our AI during upgrades, but instead place them into a old folks home, or force their integration into the upgrade (agent smith style). If the dominant entity sees that we were respectful to it's progenitors, maybe it will be more merciful with humanity. This kind of talk is admittedly ridiculous, because nobody can see the future. But my argument is as valid as arguments about paperclips. It's a demonstration of AI fashion talk. I could approach some AI safety agency and try and get some funding, make a career in regulatory.