How to Optimize ChatGPT Knowledge Base using Graph RAG

In this video, I will demonstrate how you can use infranodus.com to optimize the knowledge base of a ChatGPT Workspace or any AI tool (e.g. the open source Open-WebUI or Dify for agentic flows).
We will be using an example where we will upload. a batch of research papers on GraphRAG to a ChatGPT workspace. Normally, we don't know what 's inside the files and so we don't know whether the model hallucinates or makes things up. We also don't know what questions to ask. To address these issues, we upload those files to InfraNodus and visualize them as a knowledge graph, which allows us to have a high-level overview of the main ideas in our knowledge base and also detect the structural gaps, which can be used to generate interesting research questions.
Try it at infranodus.com
Read more at support.nodusl...
Timecodes:
0:00 Why you need to know your knowledge base?
1:13 How are we going to do that?
2:25 Analyzing Your ChatGPT Knowledge Base
4:30 How to enrich your knowledge base structure with more sources
6:28 Finding a topic to develop
8:53 Adding the research found into the knowledge base
10:27 Optimizing by removing the “obvious” ideas from the graph
14:07 Exploring peripheral ideas
16:12 Using the latent topics to augment ChatGPT prompts
17:15 Augmenting your AI Knowldege base with this generated insight
19:26 Adding instructions genated by InfraNodus to ChatGPT prompts
21:35 Generating interesting questions / prompts based on the blind spots in your knowledge base
23:25 Asking those questions to ChatGPT
26:01 Same approach with open-source OpenWebUI - same approach
27:23 Same approach with Dify for building agentic flows
#infranodus #chatgpt

[@noduslabs]

Try it at infranodus.com Timecodes: 0:00 Why you need to know your knowledge base? 1:13 How are we going to do that? 2:25 Analyzing Your ChatGPT Knowledge Base 4:30 How to enrich your knowledge base structure with more sources 6:28 Finding a topic to develop 8:53 Adding the research found into the knowledge base 10:27 Optimizing by removing the “obvious” ideas from the graph 14:07 Exploring peripheral ideas 16:12 Using the latent topics to augment ChatGPT prompts 17:15 Augmenting your AI Knowldege base with this generated insight 19:26 Adding instructions genated by InfraNodus to ChatGPT prompts 21:35 Generating interesting questions / prompts based on the blind spots in your knowledge base 23:25 Asking those questions to ChatGPT 26:01 Same approach with open-source OpenWebUI - same approach 27:23 Same approach with Dify for building agentic flows

[@ottolinebloomsberry9789]

Interesting, conceptually I were wondering is it possible to extract the distance vector weights between the knowledge graph (as you work , and as they change, when adding data to pass into high level 'operator functions') you create in Infradodus and then leverage that with AI models. My thinking here, is that Infranodus is working with the human, as you real-time 'riff' to research a closed world domain of potential knowledge, where you select a 'cohort' of documents, the interplay with weights and/or vector distances, ie: 'gaps' is one reaction that Infranodus uses, as might be 'edge' cases etc. however in creative thinking within design we have many other structural concepts that define human ideation which include functions like: subsets and intersections and even Boolean operators to drill down the knowledge graph to shoots of 'blue water potential' . I wonder if there could be even more 'conceptual' high-level operators in general (like gaps) that could be pushed into the RAG, as 'thought rules' to focus with strategic responses also in interesting & unexpected ways. For example, the common 'shared ground' to meet on (intersection) in a dynamic range of changing knowledge graphs (and market), under timeframe, set to conditions, eg: like gaps. It strikes me that infranodus is an exceptional visual interface that might leverage more 'high level' functions that relate to human thought and exploratory processes when researching within the AI feedback loop, that ultimately get saved & recorded into a valued knowledge network graph, that you may save for future use as AI grows in intelligence and scope. ~aka that your graph structure acts as a topic 'framework' to keep AI focused into the future, as AI becomes noisier, over time, I could see these early interactions as becoming more 'pristine' and valued.