Summary of Novo: Norm Voting Off Hallucinations with Attention Heads in Large Language Models, by Zheng Yi Ho et al.
NoVo: Norm Voting off Hallucinations with Attention Heads in Large Language Models
by Zheng Yi Ho, Siyuan Liang, Sen Zhang, Yibing Zhan, Dacheng Tao
First submitted to arxiv on: 11 Oct 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 The paper introduces a lightweight method called Norm Voting (NoVo) that improves the factual accuracy of large language models (LLMs) in zero-shot multiple-choice questions (MCQs). NoVo uses attention head norms to select truth-correlated head norms, which are then employed in a simple voting algorithm. This approach achieves significant gains in prediction accuracy and outperforms current state-of-the-art methods on the TruthfulQA MC1 dataset by at least 19 points. Additionally, NoVo demonstrates excellent generalization to diverse datasets, with significant gains in over 90% of them. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new method called Norm Voting (NoVo) that helps large language models (LLMs) tell the truth more often. Right now, LLMs sometimes make things up, which is bad when you need accurate answers. NoVo uses special norms to help the model decide what’s true and what’s not. This works really well and beats other methods at telling the difference between real and made-up facts. It also does a good job on many different kinds of questions. |
Keywords
» Artificial intelligence » Attention » Generalization » Zero shot
