Summary of Model Tells Itself Where to Attend: Faithfulness Meets Automatic Attention Steering, by Qingru Zhang et al.
Model Tells Itself Where to Attend: Faithfulness Meets Automatic Attention Steering
by Qingru Zhang, Xiaodong Yu, Chandan Singh, Xiaodong Liu, Liyuan Liu, Jianfeng Gao, Tuo Zhao, Dan Roth, Hao Cheng
First submitted to arxiv on: 16 Sep 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 proposes a method called AutoPASTA to improve the faithfulness of large language models (LLMs) in understanding and utilizing input contexts. LLMs often struggle with comprehending long or distracting contexts, leading to unfaithful or hallucinated responses. Current prompting methods can implicitly highlight key information but are limited in fully steering the model’s attention. AutoPASTA automatically identifies essential contextual information and explicitly highlights it by adjusting an LLM’s attention scores. The method is applied at inference time and does not require modifying any model parameters. Experimental results on open-book QA show that AutoPASTA leads to a significant improvement of 7.95% in faithfulness and performance for LLAMA3-70B-Instruct. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper helps large language models understand contexts better by automatically identifying important information and highlighting it. Right now, these models can struggle with long or distracting text, which makes their answers not very reliable. The authors propose a new method called AutoPASTA to fix this issue. It’s applied at the end of the thinking process, so the model doesn’t need any changes. By doing this, AutoPASTA helps the model understand what’s important and gives better answers. |
Keywords
» Artificial intelligence » Attention » Inference » Prompting
