This Simple Folder Beats Vector Databases. Here's Why.

The AI How · Beginner ·🔍 RAG & Vector Search ·1w ago

About this lesson

For two years, the entire AI industry agreed on how to give a model memory - chop documents into chunks, turn them into vectors, store them in a database. Then Andrej Karpathy posted a single file to GitHub and said: just use a folder of text files. It worked better. Now Google has turned that folder into an official standard. This is the clearest breakdown of the Open Knowledge Format (OKF), why it beats RAG, and the three catches nobody is talking about. ⏱ CHAPTERS 0:00 The industry just changed its mind 0:29 Why RAG existed in the first place 0:49 How RAG works — and why it never remembers 1:21 Karpathy's LLM Wiki (the idea that started it) 1:45 How the wiki works — AI writes it, not you 2:16 Google's Open Knowledge Format spec 2:48 The forgiven spec — damage control with a nicer name 3:16 Three reasons the folder beats RAG 4:11 Two quick mixups (MCP vs OKF, SEO myths) 4:29 Catch 1 — who actually tends the garden? 4:54 Catch 2 — AI breaks its own format 5:16 Catch 3 — the type field is free text 5:38 The invisible moat (this part is underrated) 6:23 Will OKF stick, or is it the next Google graveyard? 🔗 LINKS ► Andrej Karpathy's LLM Wiki gist (April 4, 2026): https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f ► OKF Spec (GitHub): https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md ► OKF Repo (GoogleCloudPlatform/knowledge-catalog): https://github.com/GoogleCloudPlatform/knowledge-catalog/tree/main/okf ► Google Cloud blog — OKF announcement: https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing 📬 NEWSLETTER Weekly breakdowns for AI builders — theaihow.com 🔔 Subscribe for takes like this every week. --- #OKF #OpenKnowledgeFormat #AIEngineering #LLMWiki #Karpathy #RAG #VectorDatabase #AIAgents #GoogleCloud #ContextEngineering #AIBuilder #LLM #AIMemory #AIArchitecture

Original Description

For two years, the entire AI industry agreed on how to give a model memory - chop documents into chunks, turn them into vectors, store them in a database. Then Andrej Karpathy posted a single file to GitHub and said: just use a folder of text files. It worked better. Now Google has turned that folder into an official standard. This is the clearest breakdown of the Open Knowledge Format (OKF), why it beats RAG, and the three catches nobody is talking about. ⏱ CHAPTERS 0:00 The industry just changed its mind 0:29 Why RAG existed in the first place 0:49 How RAG works — and why it never remembers 1:21 Karpathy's LLM Wiki (the idea that started it) 1:45 How the wiki works — AI writes it, not you 2:16 Google's Open Knowledge Format spec 2:48 The forgiven spec — damage control with a nicer name 3:16 Three reasons the folder beats RAG 4:11 Two quick mixups (MCP vs OKF, SEO myths) 4:29 Catch 1 — who actually tends the garden? 4:54 Catch 2 — AI breaks its own format 5:16 Catch 3 — the type field is free text 5:38 The invisible moat (this part is underrated) 6:23 Will OKF stick, or is it the next Google graveyard? 🔗 LINKS ► Andrej Karpathy's LLM Wiki gist (April 4, 2026): https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f ► OKF Spec (GitHub): https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md ► OKF Repo (GoogleCloudPlatform/knowledge-catalog): https://github.com/GoogleCloudPlatform/knowledge-catalog/tree/main/okf ► Google Cloud blog — OKF announcement: https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing 📬 NEWSLETTER Weekly breakdowns for AI builders — theaihow.com 🔔 Subscribe for takes like this every week. --- #OKF #OpenKnowledgeFormat #AIEngineering #LLMWiki #Karpathy #RAG #VectorDatabase #AIAgents #GoogleCloud #ContextEngineering #AIBuilder #LLM #AIMemory #AIArchitecture

Watch on YouTube ↗ (saves to browser)

Sign in to unlock AI tutor explanation · ⚡30

Related Reads

Enterprise RAG Isn’t Just About Retrieval

Learn how Retrieval-Augmented Generation (RAG) can be applied beyond retrieval to build effective enterprise AI applications

Making RAG Accurate in Python

Improve RAG accuracy in Python using hybrid dense+BM25 search and evaluation metrics like recall@k and precision@k

Dev.to · Puneet Gupta

Making RAG Accurate in Java

Improve RAG accuracy in Java using hybrid dense+BM25 search and evaluation metrics like recall@k and precision@k

Dev.to · Puneet Gupta

RAG From Scratch in Python

Learn to build Retrieval-Augmented Generation (RAG) from scratch in Python and understand its benefits over larger context windows

Dev.to · Puneet Gupta

Chapters (14)

The industry just changed its mind

0:29 Why RAG existed in the first place

0:49 How RAG works — and why it never remembers

1:21 Karpathy's LLM Wiki (the idea that started it)

1:45 How the wiki works — AI writes it, not you

2:16 Google's Open Knowledge Format spec

2:48 The forgiven spec — damage control with a nicer name

3:16 Three reasons the folder beats RAG

4:11 Two quick mixups (MCP vs OKF, SEO myths)

4:29 Catch 1 — who actually tends the garden?

4:54 Catch 2 — AI breaks its own format

5:16 Catch 3 — the type field is free text

5:38 The invisible moat (this part is underrated)

6:23 Will OKF stick, or is it the next Google graveyard?

This FREE Tool Turns ANY PDF into Perfect Markdown (MinerU Live Test)

Prompt Engineer