Summary of The Effect Of Model Size on Llm Post-hoc Explainability Via Lime, by Henning Heyen et al.
The Effect of Model Size on LLM Post-hoc Explainability via LIME
by Henning Heyen, Amy Widdicombe, Noah Y. Siegel, Maria Perez-Ortiz, Philip Treleaven
First submitted to arxiv on: 8 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 The abstract discusses how large language models (LLMs) become more performant with increased size, but there’s a gap in understanding how explainability is affected by this trend. Specifically, the authors investigate LIME explanations for DeBERTaV3 models of varying sizes on natural language inference (NLI) and zero-shot classification (ZSC) tasks. They evaluate these explanations based on faithfulness to internal decision processes and plausibility, aligning with human explanations. The key finding is that larger models do not necessarily lead to more plausible explanations despite improved performance, indicating a disconnection between LIME explanations and the models’ internal workings as size increases. Furthermore, the study highlights limitations in NLI contexts regarding faithfulness metrics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are getting bigger, but what happens to their ability to explain themselves? This paper looks at how well smaller and larger DeBERTaV3 models can be explained using LIME (Local Interpretable Model-agnostic Explanations) on two tasks: understanding sentences and predicting what they mean. The researchers want to know if these explanations are faithful to the model’s own thinking and make sense to humans. They found that bigger models don’t necessarily create more believable explanations, even though they do better at the task. This raises questions about how well we can understand how these large language models work. |
Keywords
» Artificial intelligence » Classification » Inference » Zero shot
