This is probably something that comes after Text File Attachments, but support for RAG so that the LLM can get the relevant content only when passed a very large text file or PDF would be useful.
Documentation context: rag of documentations using a website crawler (add the url of the docs) takes in all the context and you can chat with it
Gray Tomato
This is SOTA for RAG at the moment in my opinion - allows indexing of input data and creation of a searchable knowledge graph for enhanced retrieval. It should be possible to create a database that can be built by any of the available cheap models (like 2.5 flash) and queried by any available model. This creates a knowledge graph which can be reasoned over, and allows for much more integrated use of large databases. This is several generations after microsoft’s GraphRAG→nanographRAG>lightRAG>pathRAG. Each optimization makes knowledge graph construction cheaper, and the resulting graph more powerful with better question answering over the provided data.
Would be cool to have maybe an upload section in your profile to upload your llms.txt files of applications you code in to that you can ensure the code is using latest docs. Hopefully this get more upvotes.
An example of this is svelte being updated in the last year but all llms are still using old syntax and so it keeps rewriting my code incorrectly.
Log in to comment and vote
Comments3
Rose Vine
Documentation context: rag of documentations using a website crawler (add the url of the docs) takes in all the context and you can chat with it
Gray Tomato
This is SOTA for RAG at the moment in my opinion - allows indexing of input data and creation of a searchable knowledge graph for enhanced retrieval. It should be possible to create a database that can be built by any of the available cheap models (like 2.5 flash) and queried by any available model. This creates a knowledge graph which can be reasoned over, and allows for much more integrated use of large databases. This is several generations after microsoft’s GraphRAG→nanographRAG>lightRAG>pathRAG. Each optimization makes knowledge graph construction cheaper, and the resulting graph more powerful with better question answering over the provided data.
https://github.com/BUPT-GAMMA/PathRAG
https://github.com/r3d91ll/HADES-PathRAG -here’s a more recent experimental fork
Sapphire Eclipse
Would be cool to have maybe an upload section in your profile to upload your llms.txt files of applications you code in to that you can ensure the code is using latest docs. Hopefully this get more upvotes.
An example of this is svelte being updated in the last year but all llms are still using old syntax and so it keeps rewriting my code incorrectly.