factual
ChatGPT already refuses many instructions
ChatGPT already embodies a compromise on moral restraint: it will refuse plainly harmful instructions like producing a deadly virus, which most accept as good, but it also refuses benign requests like writing an unpleasant or hateful story, which Bloom argues is none of its business.
factualpending
Speaker
Paul BloomEvidence Quote
“I don't want it to tell me not to write a story that's too unpleasant because the unpleasantness makes the world worse. It's none of its business.”
Created: 6/13/2026, 12:23:39 AM
Your notes
Loading notes...