Papers with retrieve-then-generate pipelines
Dr ChatGPT tell me what I want to hear: How different prompts impact health answer correctness (2023.emnlp-main)
Copied to clipboard
| Challenge: | Using the TREC Misinformation dataset, we empirically evaluate ChatGPT to show not just its effectiveness but reveal that knowledge passed in the prompt can bias the model to the detriment of answer correctness. |
| Approach: | They empirically evaluate ChatGPT to find out whether a prompt can bias the model to the detriment of answer correctness. |
| Outcome: | The proposed model can be biased to the detriment of answer correctness by using retrieved-then-generate pipelines and how a user phrases their question as well as the question type. |