Code for Bias experiments?

[Rijgersberg]

First of all, I would like to compliment you for running bias-related experiments on the CLIP model and giving the results a prominent place in the article, not tucked away in the appendix. They are quite useful and inspire further study.

Would you be willing to release the code used to perform the bias experiments? I would like to replicate your results and expand on them.

In particular, I would like to assess:

• the effect of using the smaller models on the results of the bias experiments
• the sensitivity to the type of classes used
• the sensitivity to the prompts used for each class

If you are unable to release the code used for the experiments, would you instead be able to release the exact prompts used for the experiments?