Summary of Evaluating Gender Bias Transfer Between Pre-trained and Prompt-adapted Language Models, by Natalie Mackraz et al.
Evaluating Gender Bias Transfer between Pre-trained and Prompt-Adapted Language Models
by Natalie Mackraz, Nivedha Sivakumar, Samira Khorshidi, Krishna Patel, Barry-John Theobald, Luca Zappella, Nicholas Apostoloff
First submitted to arxiv on: 4 Dec 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper investigates how large language models (LLMs) behave when adapted for specific tasks through prompting, which is a common approach for deploying these models in real-world decision systems. The authors study the bias transfer hypothesis, which suggests that biases present in pre-trained LLMs are transferred to fine-tuned models. They find that the biases in pre-trained models like Mistral, Falcon, and Llama are strongly correlated with biases when these models are prompted for specific tasks. Furthermore, they show that this correlation remains even when prompting is used to elicit fair or biased behavior from the models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can be used for many different tasks, but they can also be unfair and discriminatory if not trained properly. This paper looks at how these models behave when we ask them to do specific things, like recognize the meaning of certain words. The researchers found that even if we ask the model to be fair or biased, it still acts in an unfair way because of its own biases. This is important to know so we can make sure these models don’t perpetuate discrimination. |
Keywords
» Artificial intelligence » Llama » Prompting
