Tenobrus argues developers must manually verify AI-generated code because models produce convincing descriptions of incorrect implementations · Digg