Summary of Famicom: Further Demystifying Prompts For Language Models with Task-agnostic Performance Estimation, by Bangzheng Li et al.
FamiCom: Further Demystifying Prompts for Language Models with Task-Agnostic Performance Estimation
by Bangzheng Li, Ben Zhou, Xingyu Fu, Fei Wang, Dan Roth, Muhao Chen
First submitted to arxiv on: 17 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 The proposed work investigates the mechanisms behind language models’ impressive performance on downstream tasks when given input prompts. Existing approaches have used label-agnostic prompt metrics like perplexity to estimate end-task performances, but these metrics are limited in their ability to accurately predict performance in complex situations like task or domain transferring scenarios. The authors propose a revised measure called FamiCom that combines familiarity with complexity, an important factor missing from current metrics. FamiCom strongly correlates with end-task performances, producing a Spearman’s correlation of 0.85 compared to 0.43 for familiarity-only metrics. The authors also apply FamiCom to automatic prompt and demonstration selection, outperforming existing methods and baselines by more than 7.0% in accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper explores how language models learn from prompts and perform better on downstream tasks. Current methods use perplexity to measure a model’s familiarity with a prompt, but this doesn’t work well for complex situations. The authors create a new metric called FamiCom that combines familiarity with the difficulty of the end task. This helps predict how well the model will do on different tasks and domains. The results show that FamiCom is much better at predicting performance than existing methods. |
Keywords
» Artificial intelligence » Perplexity » Prompt
