Lost in the Middle: How Language Models use Long Context - Explained!

Hey everyone! Thank you so much for watching this explanation of "Lost in the Middle: How Language Models use Long Context" from Liu et al. 2023! This paper has really interesting findings with regards to the impact of search quality on the resulting generation ability in RAG (Retrieval-Augmented Generation) stacks! The authors find this u-shaped curve that shows the LLM is unable to correctly answer questions when the relevant information is in the middle of the search results! This video will dive into all the experimental details, I hope you find it useful! As always, we are more than happy to answer any questions or discuss any ideas you have about the content! Paper Link: https://arxiv.org/pdf/2307.03172.pdf AutoCut in Weaviate: https://weaviate.io/developers/weaviate/search/similarity#autocut Re-Rankers in Weaviate! Re-Ranker Cohere: https://weaviate.io/developers/weaviate/modules/retriever-vectorizer-modules/reranker-cohere Re-Ranker Transformers: https://weaviate.io/developers/weaviate/modules/retriever-vectorizer-modules/reranker-transformers Hybrid Search in Weaviate: https://weaviate.io/developers/weaviate/search/hybrid Chapters 0:00 Introduction 1:18 RAG - Quick Background 3:01 Key Controls and Finding 4:05 Multi-Document QA Experiment 9:20 More Realistic Test 11:07 Key-Value Retrieval 13:13 Further Investigations 17:00 My Takeaways and New Weaviate Search Features