Writing a Book with a Knowledge Graph and AI | InfraNodus Tutorial

How to write a book using infranodus.com - presenting the workflow using my own book as an example. From a general overview to jumping into the specific parts, identifying the blind spots, and using AI to generate new ideas that bridge those gaps.
Timecodes:
00:03:15 - Discussion on growth, saturation, and ecological equilibrium, emphasizing the need for better connection of these ideas.
00:06:23 - Exploring the idea generation feature of AI, focusing on connecting disparate concepts for new insights.
00:08:23 - Mention of physical practice relating to natural cycles, suggesting a deeper exploration of this connection in the book.
00:11:00 - Analysis of terms like waves and multiple centers, and the strategic hiding of terms to reveal underlying ideas.
00:12:16 - Insight into the structural graph's ability to show narrative evolution and the importance of advanced mode for deeper analysis.
00:14:11 - Focus on detailing practices and patterns, and the reminder to delve into physical practices as a theme.
00:15:18 - Discussion on semantic variability and its optimal structure for connecting distinct topics within the discourse."

[@shawnfromportland]

the narrative influence graph is insane, I've never seen anything like that anywhere

[@noduslabs]

Glad you like it! I want to integrate it better into the Trends panel. It def deserves more attention.

[@alexcessy]

I really like the idea, and thank you for sharing. I am not trying to bash the product but I signed up and found the website essentially unusable. Could not upload even a few words to analyse. Happy to discuss further.

[@noduslabs]

Well, there are thousands of people who manage to use the tool. What do you mean you couldn't upload even a few words to analyze? Could you please provide more details?

[@preyenperumallable]

Amazing. I am about to prepare a presentation to my colleagues to share the knowledge and hopefully adopt this as a common tool .

[@noduslabs]

Great, please, let me know how it goes!

[@preyenperumallable]

@@noduslabs I will let you know, proposing a change of thinking is challenging, as I am sure you are aware. Quick question , i want to know your thoughts about making Infranodus a team sport . Would be awesome to collaborate via the knowledge graph (real time would be a perk) with a colleague in a joint session . Tried it with just me controlling the screenshare over video call and it’s not ideal. Appreciate you and team’s support hard work , it has benefitted me a lot .

[@noduslabs]

@@preyenperumallable Thank you! In fact there is a collaboration mode, but I'm changing it now, so soon it'll be available again. For now I think it would work if you just log in using the same account and open the same graph. Have you tried it?

[@preyenperumallable]

Exciting to hear about the possibility of collaboration features ! I haven’t tried it since I’m using a premium account out of my own pocket and I’m trying to convince the others to do the same . For now I will just show them how awesome a tool it can be by showing them my outputs. I’m a medical device design engineer so the use cases already are fascinating (at least to me) as well as my fascination with jazz improvisation. (PS if someone scrapes these comments one day in their Infranodus to put into their knowledge graph: Hello !)😂

[@noduslabs]

​@@preyenperumallable Ahha very interesting! Maybe one day those comments will end up in InfraNodus, so I'm saying hello as well here :) I think that jazz improvisation has a lot to do with InfraNodus because I find the moments I get the most out of it is when I'm using it as a musical instruments (I'm kind of into techno / bass music jazz-like vibes myself). So it makes a lot of sense to me.

[@pavlophoenix]

thats what im looking for . how to connect infranodus to Obsidian?! and i like thath theres could be like a big all knowing personnaly trained by you , AGI agent (like Jarvis) and thats what i need for learning stuff so my head doesnt have to keep this thougts in . Can it be real?

[@noduslabs]

For now, you can import your Obsidian docs to InfraNodus but for a live integration you have to wait a month or two. We're releasing it soon. Thanks!

[@mauricioacosta7492]

Could you use llama instead of gpt? and make it local?

[@noduslabs]

Not yet, but we're working on it. What would be your use case to make it local?

[@Tymon0000]

@@noduslabs maybe to not feed your data to third parties...

[@mauricioacosta7492]

@@noduslabs Running our business with AI assistance locally independent from the internet. Also it would be totally free if we used llama right?

[@DmitryParanyushkin]

@@Tymon0000well when you use gpt 4 they don’t train their models on it

[@noduslabs]

@@mauricioacosta7492no it wouldn’t be totally free because I’d still want to get paid for the work I’ve done developing it :)

[@jonmichaelgalindo]

Does co-occurence really represent semantic connectedness?

[@DmitryParanyushkin]

It’s even more interesting. It represents connectedness in this particular text. So you understand the context much better.

[@jonmichaelgalindo]

@@DmitryParanyushkin In a murder mystery, the clues are all "connected" to the murder; but they do not co-occur. In a fantasy novel, the kingdoms and cities are connected by characters' expressed geopolitical views, but they do not co-occur. In the hero's journey, each trial is connected to the others through the hero's growth, but not via co-occurence. The subtlety and challenge of discovering the connectedness within a story is one of the joys of reading. I don't see what it has to do with co-occurence though.

[@noduslabs]

It has a lot to do with co-occurrence, because in the context of your particular comment, for instance, I can say that you used the term "geopolitical" next to the "kingdom" and "cities" and "characters" - so that gives me a good understand of what you're talking about. When you build a graph this way from any text, you can identify the main ideas, the pathways for meaning circulations, and clusters. So it's pretty informative.

[@jonmichaelgalindo]

@@noduslabs You only knew those things were connected because you are intelligent. Hero also co-occured with geopolitical (by proximity), but not by connectedness. IBM tried this with Watson, right? And then they realized that the beginnings and ends of chunks of text were also connected. (In your comment, the word "it" occurs at the beginning, and again at the end, and the two are implicitly connected through that beginning/end placement.) Then we got auto-encoders and vector databases, which is the most similar to this (although vastly more powerful). But that was still way too far from measuring connectedness the way you just did for goepolitical/kingdom/etc. Transformer architectures got us one step closer, and now AI can do really impressive things. But co-occurence is just not meaningfully useful from what I understand. Not that I have any problem with it.

[@noduslabs]

​@@jonmichaelgalindo It's interesting, because I think that vectors and RAGs are overhyped. People talk about them like it's something more advanced. I don't think so. Let's have this discussion, I wonder how you will respond to this. The way I see it, a vector is basically a representation of how a certain word is embedded into language in general. In simple terms: what other words it likes to hang out with (or tokens, or phrases). However, when I want to analyze a specific text and understand its distinct features I actually don't want to know generalities, I want to know how particular words like to hang out within the context of this text, so I'm not interested in this general embedding, I want to know how those words like to hang out in this specific text. That's why co-occurrence is a very simple way to understand that. Another advantage is that it mimicks our perception. We usually have a 10-second gap where we're aware of what we're talking about. Like if I ask you now what exactly I said 10 seconds ago, you'll have a harder time remembering it than if I ask you what I'm talking about right now. So that co-occurrence representation is in fact very useful, because it represents something that's very much connected to our perception.