Both parties take some time to reach some mutual understanding of what the architecture is. Or at least build some intuition about what kind of architecture the other has in mind or what changes to perform in that architecture. This is a common pattern that I have seen a few times. I have heard that this is called typical in pre-paradigmatic fields. There is not enough standard terminology for common patterns or constraints. Or alternatively, there are not enough concrete implementations to point to or to be able to write down quickly. It would have been much easier if johnswentworth had been able to refer to a “Shah representation of the concept of human value” or a “the 2nd order Byrnes search process neuron hull” (all made up) and say “let’s Yud-chain the Byrnes hull to the Shah”. And then the discussion would have been whether the 2nd order Byrnes converges faster than the Yud-chain diverges or something. But we are not there and it shows.
Both parties take some time to reach some mutual understanding of what the architecture is. Or at least build some intuition about what kind of architecture the other has in mind or what changes to perform in that architecture. This is a common pattern that I have seen a few times. I have heard that this is called typical in pre-paradigmatic fields. There is not enough standard terminology for common patterns or constraints. Or alternatively, there are not enough concrete implementations to point to or to be able to write down quickly. It would have been much easier if johnswentworth had been able to refer to a “Shah representation of the concept of human value” or a “the 2nd order Byrnes search process neuron hull” (all made up) and say “let’s Yud-chain the Byrnes hull to the Shah”. And then the discussion would have been whether the 2nd order Byrnes converges faster than the Yud-chain diverges or something. But we are not there and it shows.