Summary of Properties and Challenges Of Llm-generated Explanations, by Jenny Kunz et al.
Properties and Challenges of LLM-Generated Explanations
by Jenny Kunz, Marco Kuhlmann
First submitted to arxiv on: 16 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC); 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 the capabilities of large language models (LLMs) to generate explanations for their outputs in various domains. While LLMs are trained on specific data sets, they often provide explanations without explicit annotation. The study finds that generated explanations from multi-domain instruction fine-tuning data exhibit selectivity and illustrative elements, but less frequently contain subjective or misleading information. The authors discuss the implications of these properties for self-rationalising systems, highlighting both positive and negative aspects depending on the goals and user groups. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research looks at how big language models explain their answers to questions. These models are trained on a lot of data, but they often provide explanations even if there’s no specific instruction to do so. The study finds that these explanations tend to be focused and include helpful examples, but not always biased or misleading. The authors think about what this means for systems that try to explain their own actions, pointing out both good and bad aspects depending on the goals and people using them. |
Keywords
* Artificial intelligence * Fine tuning
