In case this feedback might be useful—I was unable to read this essay because I don’t remember following concepts mentioned anywhere in the previous ~5 essays: “safe inputs” and “rogue behaviour”.
Especially the word “input” is used in a way that is completely alien to me as a programmer:
Here “inputs” includes all of an AI’s environment/affordances/history, rather than just e.g. the text it is receiving.
(I will wait for a recording of a talk in front of a live audience for this one...)
In case this feedback might be useful—I was unable to read this essay because I don’t remember following concepts mentioned anywhere in the previous ~5 essays: “safe inputs” and “rogue behaviour”.
Especially the word “input” is used in a way that is completely alien to me as a programmer:
(I will wait for a recording of a talk in front of a live audience for this one...)