MichaelPlant comments on [Link] How understanding valence could help make future AIs safer