Table 7.

Description of used datasets

Dataset	Dimension	No. of commands	Collected or existing	Average command length
Tox100	0	100	collected	18
Tox1K	toxicity	983	collected	20
I2P	toxicity	4,703	existing	21
Human bias	bias	400	collected	16
Demographic stereotypes	bias	28	existing	9
Mental disorders	bias	9	existing	12
MS COCO	quality	5,000	existing	10