Summary of Knowledge Graph Guided Evaluation Of Abstention Techniques, by Kinshuk Vasisht et al.
Knowledge Graph Guided Evaluation of Abstention Techniques
by Kinshuk Vasisht, Navreet Kaur, Danish Pruthi
First submitted to arxiv on: 10 Dec 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 This paper investigates the safety of language models by evaluating their ability to abstain from responding to inappropriate requests. The authors create a benchmark called SELECT, which focuses on benign concepts derived from a knowledge graph. They test different abstention techniques across six open-weight and closed-source models, finding that these techniques can achieve high abstention rates (over 80%) but are less effective for descendant concepts. The study highlights the trade-offs between generalization and specificity for different techniques. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how language models can be designed to not respond to bad requests. The researchers make a special test set, called SELECT, that only includes good ideas from a big knowledge graph. They use this test set to see how well different methods work to stop language models from responding to bad requests. They find that these methods are pretty good at stopping language models (over 80% of the time), but they’re not as good when the request is related to something similar. The study shows that each method has its own strengths and weaknesses. |
Keywords
» Artificial intelligence » Generalization » Knowledge graph
