#artificialintelligence #132:? LLMs and Graph neural networks - an overview and impact

In the last edition of this newsletter, I asked the question: Can a chatbot replace a research paper?

A comment on that post by Anthony Alcaraz captures the vision accurately

This is a very interesting idea :

The idea of publishing "agents" instead of traditional papers is interesting. Some potential benefits I see:

Allows for interactive dissemination of research, versus static papers.

Could enable automated peer review by other agents.

Provides a way to encapsulate research in an accessible format.

You could also open a format where you could associate different agents for papers related.

AutoGen library by Microsoft allow to build such system relatively easily.

I m sure there is a Revolution in scientific knowledge sharing to open up here

The ideas presented in that post are based on the concept of:?

1. Rethinking education from first principles using AI.?
2. Amplifying human intellect
3. Implemented using knowledge graphs and LLMs

Amplifying human knowledge and intellect can take many forms: ex an agent which represents a book, a paper, an expert etc

In this post - I discuss how this idea could be implemented using knowledge graphs and LLMs

I have been developing this idea in my teaching and also my start-ups (Salooki and Erdos)

We are also working with some leading thinkers in this area

The first part starts with developing an educational chatbot using LLMs and knowledge graphs using neo4j - which was shared by David Stevens ?

If you want to explore further, see these three blogs

• Harnessing LLMs With Neo4j
• Fine-Tuning vs Retrieval-Augmented Generation
• Multi-Hop Question Answering

And GitHub repository,?

We are also looking at knowledge graphs and LLMs - based on the work of Jerry Liu

Wenqi Glantz has also been contributing to this work in her excellent blog? 7 Query Strategies for Navigating Knowledge Graphs With LlamaIndex?

If you want to see this strategy holistically, see the work of Anthony Alcaraz who says that knowledge graphs and embeddings are two sides of the same coin

The same idea is expressed by Tony Seale in Vectors need graphs

Vectors need Graphs! Embedding vectors are a pivotal tool when using Generative AI. While vectors might initially seem an unlikely partner to graphs, their relationship is more intricate than it first appears.

Collectively, this gives us a good view of how LLMs and knowledge graphs connect together

If you are interested in this area, please see our course at the #universityofoxford on #artificialintelligence - Artificial Intelligence: Generative AI, Cloud and MLOps (online)

UPDATE

Just saw this great post from Boris Villazon-Terrazas, PhD - Blending LLMs with Knowledge Graphs - h/t Rania Bayoumy

Images source: Adam Cowley (midjourney) https://neo4j.com/developer-blog/building-educational-chatbot-neo4j/