Summary of The Probabilities Also Matter: a More Faithful Metric For Faithfulness Of Free-text Explanations in Large Language Models, by Noah Y. Siegel et al.
The Probabilities Also Matter: A More Faithful Metric for Faithfulness of Free-Text Explanations in Large Language Models
by Noah Y. Siegel, Oana-Maria Camburu, Nicolas Heess, Maria Perez-Ortiz
First submitted to arxiv on: 4 Apr 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 approach to assessing the faithfulness of natural language explanations provided by large language models (LLMs) is introduced in this work. The authors propose Correlational Explanatory Faithfulness (CEF), a metric that takes into account the total shift in the model’s predicted label distribution, offering a more accurate reflection of the explanations’ faithfulness. This metric is used to develop the Correlational Counterfactual Test (CCT) which is then instantiated on the Counterfactual Test (CT) from Atanasova et al. (2023). The authors evaluate the faithfulness of free-text explanations generated by few-shot-prompted LLMs from the Llama2 family on three NLP tasks and find that their metric measures aspects of faithfulness which the CT misses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can provide natural language explanations, but it’s unclear how faithful these explanations are. This paper introduces a new way to measure the faithfulness of these explanations by looking at how much they change when we change the input. The authors use this method to test whether the explanations generated by some big language models are accurate or not. They find that their method is better than previous methods at measuring faithfulness, and it can help us understand how AI systems make decisions. |
Keywords
» Artificial intelligence » Few shot » Nlp
