The hottest Substack posts of
Vigneshwarar’s Newsletter
And their main takeaways
- The current HackerNews ranking algorithm is based on a simple formula involving points, age, and a constant factor.
- Proposing a new approach called HackerRank that incorporates PageRank-like scoring for user profiles based on upvotes and takes flagging into account.
- Additional ideas for improving the ranking algorithm include considering user submission upvotes, reading time, and website reputation.
- RAG technique improves factual accuracy by combining LLMs with retrieved documents
- EazyRAG focuses on effective context formation by letting GPT handle it dynamically
- Manual customization of context formation can be avoided by utilizing GPT's capabilities
- Retrieval-Augmented Generation (RAG) pipeline can be built without using trendy libraries like Langchain
- RAG technique involves retrieving related documents, combining them with language models, and generating accurate information
- RAG pipeline involves data preparation, chunking, vector store, retrieval/prompt preparation, and answer generation steps
- EazyRAG simplifies the process of implementing RAG in browsers with a user-friendly API.
- Indexed content on the server side allows easy generation of grounded answers through a simple API call.
- There is support for streams and an aim to create a beginner-friendly API for LLM agents.