Summary of Local Explanations and Self-explanations For Assessing Faithfulness in Black-box Llms, by Christos Fragkathoulas et al.
Local Explanations and Self-Explanations for Assessing Faithfulness in black-box LLMs
by Christos Fragkathoulas, Odysseas S. Chlapanis
First submitted to arxiv on: 18 Sep 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 A novel task is introduced to evaluate the faithfulness of large language models (LLMs) using local perturbations and self-explanations. The proposed explainability technique inspired by leave-one-out approaches identifies sufficient and necessary parts for correct answers, serving as explanations. A metric for assessing faithfulness compares these crucial parts with model self-explanations. This approach is validated on the Natural Questions dataset, demonstrating its effectiveness in explaining model decisions and evaluating faithfulness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a new way to test how well large language models understand things by changing small parts of their answers and asking them to explain why they chose those answers. The method works by finding what parts of the answer are most important for getting it right, then comparing those parts to the model’s own explanation. This helps us figure out if the model is really understanding what it’s being asked or just lucky. The team tested this approach using a big dataset and showed that it can help us understand how well language models are doing. |
