This piece was helpful in outlining how different people in the AI safety space disagree, and what the issues with Paul’s approaches seem to be. Paul’s analogies with solving hard problems was especially interesting to me (the point where most problems don’t seem to occupy a position midway between totally impossible and solvable). The inline comments by Paul were also good to read as counterpoints to Eliezer’s responses.
This piece was helpful in outlining how different people in the AI safety space disagree, and what the issues with Paul’s approaches seem to be. Paul’s analogies with solving hard problems was especially interesting to me (the point where most problems don’t seem to occupy a position midway between totally impossible and solvable). The inline comments by Paul were also good to read as counterpoints to Eliezer’s responses.