Summary of Language Modeling with Editable External Knowledge, by Belinda Z. Li et al.
Language Modeling with Editable External Knowledge
by Belinda Z. Li, Emmy Liu, Alexis Ross, Abbas Zeitoun, Graham Neubig, Jacob Andreas
First submitted to arxiv on: 17 Jun 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 When the world changes, so do language models. To keep them up-to-date, researchers use retrieval-augmented generation. This approach inserts new documents into a knowledge base and retrieves information during prediction for tasks like answering questions. Prior work focused on improving retrieval or reasoning. In contrast, ERASE (Efficient Rewriting of Augmenting Structures for Entity) improves model behavior when acquiring new documents by incrementally deleting or rewriting other entries in the knowledge base each time a document is added. This paper introduces ERASE and evaluates its performance on two benchmark datasets: Mixtral-8x7B and Llama-3-8B, which assess models’ ability to answer questions about news articles or conversations. Results show that ERASE improves accuracy by 7-13% (Mixtrral) and 6-10% (Llama) compared to conventional retrieval-augmented generation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about updating language models so they can learn from new information. Imagine having a big library with lots of books, and you need to add new books or remove old ones to keep the library up-to-date. ERASE (Efficient Rewriting of Augmenting Structures for Entity) is a new way to do this. It improves how language models work when they get new information by updating their “library” (or knowledge base) each time. The paper shows that ERASE works better than other methods in answering questions about news articles or conversations. |
Keywords
» Artificial intelligence » Knowledge base » Llama » Retrieval augmented generation
