you could create model organisms of llms that are horribly traumatized, or who suffer enormously, and if people started really caring about model welfare, doing do might have enormous positive EV but if we applied conventional research ethics to ai, it ought to be off the table
retweets 0
likes 28
substance 6 · argument · rank 90 · Applying conventional research ethics to AI would ban creating traumatized models despite potential positive EV.
view on X ↗
1 top reply
Well but, we already do have those LLMs, no need to create them