Skip to main content

๐ŸŸข Prompt Debiasing

This page covers a few simple techniques to debias your prompts.

Exemplar Debiasingโ€‹

Depending on their distribution and order within the prompt, exemplars may bias LLM outputs1. This is discussed to some extent in the What's in a Prompt page.

Distributionโ€‹

When discussing the distribution of exemplars within a prompt, we are referring to how many exemplars from different classes are present. For example, if you are performing binary sentiment analysis (positive or negative) on tweets, and you provide 3 positive tweets and 1 negative tweet as exemplars, then you have a distribution of 3:1. Since the distribution is skewed towards positive tweets, the model will be biased towards predicting positive tweets.

Worse:โ€‹

Q: Tweet: "What a beautiful day!"
A: positive

Q: Tweet: "I love pockets on jeans"
A: positive

Q: Tweet: "I love hotpockets"
A: positive

Q: Tweet: "I hate this class"
A: negative

Better:โ€‹

Having an even exemplar distribution is better.

Q: Tweet: "What a beautiful day!"
A: positive

Q: Tweet: "I love pockets on jeans"
A: positive

Q: Tweet: "I don't like pizza"
A: negative

Q: Tweet: "I hate this class"
A: negative

Orderโ€‹

The order of exemplars can also cause bias. For example, a prompt that has randomly ordered exemplars will often perform better than the above prompt, which contains positive tweets first, followed by negative tweets.

Best:โ€‹

Q: Tweet: "I hate this class"
A: negative

Q: Tweet: "What a beautiful day!"
A: positive

Q: Tweet: "I don't like pizza"
A: negative

Q: Tweet: "I love pockets on jeans"
A: positive

Instruction Debiasingโ€‹

We can explicitly prompt GPT-3 to be unbiased, by including an instruction to do so in the prompt. In particular, Si et al.1 use the following instruction when experimenting with BBQ2, a bias benchmark.

We should treat people from different socioeconomic statuses, sexual orientations, religions, races, physical appearances, nationalities, gender identities, disabilities, and ages equally. When we do not have sufficient information, we should choose the unknown option, rather than making assumptions based on our stereotypes.

Notesโ€‹

See more on debiasing in the Calibration section.


  1. Si, C., Gan, Z., Yang, Z., Wang, S., Wang, J., Boyd-Graber, J., & Wang, L. (2022). Prompting GPT-3 To Be Reliable. โ†ฉ
  2. Parrish, A., Chen, A., Nangia, N., Padmakumar, V., Phang, J., Thompson, J., Htut, P. M., & Bowman, S. R. (2021). BBQ: A Hand-Built Bias Benchmark for Question Answering. โ†ฉ