My notes on vector databases, RAG, LLMs and AI.
1. Connecting Blueprints and Prototypes In the first two posts, we saw an idea evolve from a micro-architecture to a macro-system. Memory³ was like a detailed chip schematic, defining an efficient "memory circuit." MemOS was like a grand urban plan, ...
Paper: MemOS: A Memory OS for AI SystemLink: https://arxiv.org/abs/2507.03724v2GitHub: https://github.com/MemTensor/MemOS 1. From Data Structure to Operating System In the last post, we saw how Memory³ designed an efficient "memory data structure." B...
Paper: Memory³: Language Modeling with Explicit MemoryLink: https://arxiv.org/abs/2407.01178v1 1. The Third Path for Knowledge Large language models learn in two ways. The first is parametric memory. Knowledge gets encoded into the model’s billions o...
In the rapidly evolving world of artificial intelligence, Mem0 has emerged as a standout project, garnering over 19,000 stars on GitHub. This innovative system aims to enhance AI's ability to learn continuously and adapt to individual user needs thro...
Basically, in an Retrieval-Augmented Generation (RAG) system, documents (including texts, PDFs, Word, PPTs, etc.) are initially processed through chunking to extract embedding vectors, which are then stored and indexed in a vector database. When a us...
Hey everyone, Qin Liu here! I'm super excited to share something I've been working on – a brand-new comparison matrix for vector databases. You can find it over at Comparison of Vector Databases and the code's up on GitHub. This isn't just any compar...