Summary of From Rags to Rich Parameters: Probing How Language Models Utilize External Knowledge Over Parametric Information For Factual Queries, by Hitesh Wadhwa et al.
From RAGs to rich parameters: Probing how language models utilize external knowledge over parametric information for factual queries
by Hitesh Wadhwa, Rahul Seetharaman, Somyaa Aggarwal, Reshmi Ghosh, Samyadeep Basu, Soundararajan Srinivasan, Wenlong Zhao, Shreyas Chaudhari, Ehsan Aghazadeh
First submitted to arxiv on: 18 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 Retrieval Augmented Generation (RAG) has gained popularity for enhancing language model responses using external context. This paper delves into the RAG pipeline, revealing that language models take a shortcut by leveraging context information to answer questions, relying minimally on their parametric memory. Causal Mediation Analysis and Attention Contributions and Knockouts demonstrate this behavior in LLaMa and Phi models, showcasing a pronounced bias towards utilizing context over internal memory. This study sheds light on the mechanistic workings of RAG, highlighting its limitations and potential applications in search, question/answering, and chat-bots. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how language models work when given extra information to help answer questions. It’s like having a hint or a clue that helps you figure out the answer. The researchers found that these models don’t really use their own internal memory to answer questions, but instead rely on the external information they’re given. They used special tools to study how this works in different language models and found that it’s a common pattern across many of them. |
Keywords
» Artificial intelligence » Attention » Language model » Llama » Question answering » Rag » Retrieval augmented generation
