Timecodes: 0:00 Introduction 0:38 Importing ChatGPT conversation to InfraNodus 1:12 How InfraNodus text network analysis works 2:14 Step 1: General overview of a ChatGPT conversation and summary 3:11 Step 2: Find the blind spots in the conversation to connect new ideas 4:01 Step 3: Stimulating your thinking using built-in GPT4 AI 4:43 Step 4: Saving ideas into project notes / adding into the graph 6:16 Step 5: Finding more structural gaps in the conversation 6:55 Step 6: Feeding AI-generated content back to AI (human-in-the-loop approach) 8:05 Summary of the structural gap approach 8:53 Step 7: Peeling off the top layer of evident ideas (reveal the non-obvious) 10:20 Step 8: Developing non-evident topics in your conversation 11:06 Step 9: Modifying the graph view 11:20 Step 10: Jumping into specific parts of the texts (knowledge graph search) 12:53 Summary of the "revealing the non obvious" approach 13:13 Step 11: Finding topics in the knowledge graphs (tip: look at the periphery) 13:40 Step 12: Continuing the conversation in ChatGPT 14:03 Step 13: Using the project notes to develop your conversation 15:53 Why InfraNodus is useful for enhancing your ChatGPT chats
Наш мозг по сути работает по таким графам и мы называем это ассоциациями. Я уверен, что на основе таких графов можно улучшить способности искусственного интеллекта, если научить его самостоятельно делать такие графы и использовать не только слова, а также и видео и аудио информацию
Thank you Dmitry. I have a strategic question as I try to understand InfraNodus better. Is the product the end state or a means to an end? For example since fine tuning a LLM is generally a complex set of procedures and requires ongoing RLHF, is this product able to assist in evolution of the LLM itself or a secondary resource of these notes and documents (sort of how Obsedian docs and graph become one and the same)
Good point! I think it's both. At the moment, I focus more efforts on finding interesting applications, but, of course, I'm also working on researching how these approaches can be used to evolve LLMs. For instance, structural gaps are a really interesting way to push LLMs to be more creative and it's more precise than just adjusting the temperature.
@@noduslabs This talk distills the topic very well, especially the section “3 approaches to LLM and KG” ua-cam.com/video/1RZ5yIyz31c/v-deo.htmlsi=AOVHOqk6rtFb5OtP
“Our task here!” I had to pause the video at 4:30. You said something profound, “our task here is to stimulate our thinking.” I'm concerned that people in general will allow their Center of Creativity, their Center of Thinking, to reside in the artificial intelligence. The center of creativity, the center of thinking, the center of idea generation, needs to remain in the self - in me as I'm working. Excuse my interruption. 🙂 back to the video…
Great point! I also believe that we shouldn't outsource our thinking process to AI, rather use it for inspiration and making sure it doesn't become the only source of inspiration.
Well, at the moment you'd get all the words / concepts used. Do you mean you only want to be able to extract some terms? In this case, what you can do is to run your handbook through ChatGPT asking it to highlight the terms using [[double brackets]] syntax. Then you plug the result in InfraNodus and it will visualize connections only between those terms with the double brackets.
Interesting thank you. But I feel the video falls short of the next step which is: Once I have the idea developed to satisfaction, what outputs can I generate, or how would this then help structure a complex article?
Yes, I also missed out on how to then write a complex article, find a place to publish it, make sure it gets published, then get some feedback on it, incorporate that feedback into my work, and eventually come up with something truly amazing that would blow everyone's mind :))
@@noduslabs No seriously, I can do all you say perfectly well. But to refine my question, once I have my knowledge organized and linked in Infranodus, are there export functions, like exporting outputs based on the developed context. I would imagine doing work within the software and exporting the developed context, and then continue doing what you suggest above.
Good Day. Question. Is it possible to connect InfraNodus to a code database infrastructure like for eg: let's say someone built an App and Infranodus happens to have access to the code infrastructure and repository. Is it possible for a user to stimulate the InfraNodus application in thinking about potential ways to improve on existing code infrastrucure. Let me break it down a bit more.. I think AI tools like ChatGPT 4o is good at writing code but not in a way where it is always suitable for production level deployment. I believe Infranodus would serve well in enabling professional software developers to see how the AI develops its thinking around solving code bugs and in making improvements by working on the gaps that might exist in the neural connections in ChatGPT and other LLMs. Is it possible for Infranodus to pick up on gaps in where the LLMs might go wrong with thinking about a code logic/recommendation, make new, creative connections on the LLMs extensive knowlege base, potentially apply those lessons to the existing code infrastructure for improvement?
That would be a very interesting use case. What do you think would be best to use as nodes in the graph in this case? Function names? And how would the connections work? If a function refers to another function? Let me know, I'd be very curious!
This tool is very interesting. Would it be possible to use this tool to generate technical publications for new products? For example, as engineers design a new product and generate informal documents that describe the purpose, functionality, and operation of this new product and its components, such documentation would then be imported into InfraNodus to generate a knowledge graph. The knowledge graph would eventually have all of the information about this new product, but not all of it is relevant to all users. Could we use Infranodus to generate various publications for different users based on the information contained in the knowledge graph? For example, we would tell Infranodus to generage a User Manual for the end user, a Service Manual for a service technician, and a Training Guide for a sales person?
Great idea for a product! We can think together what would be the best way to achieve that. For instance, you, as an an engineer could select the parts of the graph you want the content to be generated from and it could use only those parts of the graph, right? Would you want it to use the original text that served as the source material for creating the graph or that the generated text is synthetic and is generated new?
@@noduslabs I currently work in the technical training and writing field. Our role is to take the highly technical and often complex information from the engineers and translate it into a format that is easily understood by the average user. I can see how Infranodus could potentially be used to help make the process more efficient. I have not used Infranodus. I have only watched a couple of your demonstration videos. So based on my limited knowledge of the tool, this is how I envision the process: 1) All available raw data and information about the new product is imported into the knowledge graph to create the main database of product information. 2) Technical writers use Infranodus to analyze the knowledge graph to identify any gaps or deficiencies and work with the engineers to fill in those gaps. 3) The technical writers and/or developers create templates for the desired technical documentation. These templates define the layout of the documents. For example, an end user guide would be formatted differently from a service manual which would be be different from a training manual. 4) The technical writers tell Infranodus/ChatGPT to generate the desired document. Infranodus/ChatGPT then takes the relevant information, formats it to fit the matching template and generates the document. 5) The technical writer reviews the document for errors and makes corrections as necessary. 6) Rather than updating the technical documentation as the product changes over time, the knowledge graph would be updated, which would then be used to update any affected technical documents. I'm not sure how feasible this idea is, but it seems feasible, which is why I reached out. I can see how such functionality would be very useful in many applications. I'd be interested to hear your feedback.
Timecodes:
0:00 Introduction
0:38 Importing ChatGPT conversation to InfraNodus
1:12 How InfraNodus text network analysis works
2:14 Step 1: General overview of a ChatGPT conversation and summary
3:11 Step 2: Find the blind spots in the conversation to connect new ideas
4:01 Step 3: Stimulating your thinking using built-in GPT4 AI
4:43 Step 4: Saving ideas into project notes / adding into the graph
6:16 Step 5: Finding more structural gaps in the conversation
6:55 Step 6: Feeding AI-generated content back to AI (human-in-the-loop approach)
8:05 Summary of the structural gap approach
8:53 Step 7: Peeling off the top layer of evident ideas (reveal the non-obvious)
10:20 Step 8: Developing non-evident topics in your conversation
11:06 Step 9: Modifying the graph view
11:20 Step 10: Jumping into specific parts of the texts (knowledge graph search)
12:53 Summary of the "revealing the non obvious" approach
13:13 Step 11: Finding topics in the knowledge graphs (tip: look at the periphery)
13:40 Step 12: Continuing the conversation in ChatGPT
14:03 Step 13: Using the project notes to develop your conversation
15:53 Why InfraNodus is useful for enhancing your ChatGPT chats
Наш мозг по сути работает по таким графам и мы называем это ассоциациями. Я уверен, что на основе таких графов можно улучшить способности искусственного интеллекта, если научить его самостоятельно делать такие графы и использовать не только слова, а также и видео и аудио информацию
This is very interesting 🤔
Thank you Dmitry. I have a strategic question as I try to understand InfraNodus better. Is the product the end state or a means to an end? For example since fine tuning a LLM is generally a complex set of procedures and requires ongoing RLHF, is this product able to assist in evolution of the LLM itself or a secondary resource of these notes and documents (sort of how Obsedian docs and graph become one and the same)
Good point! I think it's both. At the moment, I focus more efforts on finding interesting applications, but, of course, I'm also working on researching how these approaches can be used to evolve LLMs. For instance, structural gaps are a really interesting way to push LLMs to be more creative and it's more precise than just adjusting the temperature.
@@noduslabs can Infranodus be hosted on aws or run in a container?
@@noduslabs This talk distills the topic very well, especially the section “3 approaches to LLM and KG” ua-cam.com/video/1RZ5yIyz31c/v-deo.htmlsi=AOVHOqk6rtFb5OtP
@@RedCloudServices Yes, it can, but we are not offering it, because we only provide the cloud version.
I have seen many things but this is an interesting way to do it
“Our task here!” I had to pause the video at 4:30. You said something profound, “our task here is to stimulate our thinking.” I'm concerned that people in general will allow their Center of Creativity, their Center of Thinking, to reside in the artificial intelligence. The center of creativity, the center of thinking, the center of idea generation, needs to remain in the self - in me as I'm working. Excuse my interruption. 🙂 back to the video…
Great point! I also believe that we shouldn't outsource our thinking process to AI, rather use it for inspiration and making sure it doesn't become the only source of inspiration.
Great tutorial. Your image frequently covers what you write. Could you please locate your image on the right side of the screen?
Haha then it will cover the analytics panel. Yes, it's a problem. I'm going to see how we can solve it.
Would it be possible to import an economic handbook and get all the economic terms used and see the relationships between these terms?
Well, at the moment you'd get all the words / concepts used. Do you mean you only want to be able to extract some terms? In this case, what you can do is to run your handbook through ChatGPT asking it to highlight the terms using [[double brackets]] syntax. Then you plug the result in InfraNodus and it will visualize connections only between those terms with the double brackets.
Interesting thank you. But I feel the video falls short of the next step which is: Once I have the idea developed to satisfaction, what outputs can I generate, or how would this then help structure a complex article?
Yes, I also missed out on how to then write a complex article, find a place to publish it, make sure it gets published, then get some feedback on it, incorporate that feedback into my work, and eventually come up with something truly amazing that would blow everyone's mind :))
@@noduslabs No seriously, I can do all you say perfectly well. But to refine my question, once I have my knowledge organized and linked in Infranodus, are there export functions, like exporting outputs based on the developed context. I would imagine doing work within the software and exporting the developed context, and then continue doing what you suggest above.
Good Day. Question. Is it possible to connect InfraNodus to a code database infrastructure like for eg: let's say someone built an App and Infranodus happens to have access to the code infrastructure and repository. Is it possible for a user to stimulate the InfraNodus application in thinking about potential ways to improve on existing code infrastrucure. Let me break it down a bit more..
I think AI tools like ChatGPT 4o is good at writing code but not in a way where it is always suitable for production level deployment. I believe Infranodus would serve well in enabling professional software developers to see how the AI develops its thinking around solving code bugs and in making improvements by working on the gaps that might exist in the neural connections in ChatGPT and other LLMs.
Is it possible for Infranodus to pick up on gaps in where the LLMs might go wrong with thinking about a code logic/recommendation, make new, creative connections on the LLMs extensive knowlege base, potentially apply those lessons to the existing code infrastructure for improvement?
That would be a very interesting use case. What do you think would be best to use as nodes in the graph in this case? Function names? And how would the connections work? If a function refers to another function? Let me know, I'd be very curious!
This tool is very interesting. Would it be possible to use this tool to generate technical publications for new products? For example, as engineers design a new product and generate informal documents that describe the purpose, functionality, and operation of this new product and its components, such documentation would then be imported into InfraNodus to generate a knowledge graph. The knowledge graph would eventually have all of the information about this new product, but not all of it is relevant to all users. Could we use Infranodus to generate various publications for different users based on the information contained in the knowledge graph? For example, we would tell Infranodus to generage a User Manual for the end user, a Service Manual for a service technician, and a Training Guide for a sales person?
Great idea for a product! We can think together what would be the best way to achieve that. For instance, you, as an an engineer could select the parts of the graph you want the content to be generated from and it could use only those parts of the graph, right? Would you want it to use the original text that served as the source material for creating the graph or that the generated text is synthetic and is generated new?
@@noduslabs I currently work in the technical training and writing field. Our role is to take the highly technical and often complex information from the engineers and translate it into a format that is easily understood by the average user. I can see how Infranodus could potentially be used to help make the process more efficient. I have not used Infranodus. I have only watched a couple of your demonstration videos. So based on my limited knowledge of the tool, this is how I envision the process:
1) All available raw data and information about the new product is imported into the knowledge graph to create the main database of product information.
2) Technical writers use Infranodus to analyze the knowledge graph to identify any gaps or deficiencies and work with the engineers to fill in those gaps.
3) The technical writers and/or developers create templates for the desired technical documentation. These templates define the layout of the documents. For example, an end user guide would be formatted differently from a service manual which would be be different from a training manual.
4) The technical writers tell Infranodus/ChatGPT to generate the desired document. Infranodus/ChatGPT then takes the relevant information, formats it to fit the matching template and generates the document.
5) The technical writer reviews the document for errors and makes corrections as necessary.
6) Rather than updating the technical documentation as the product changes over time, the knowledge graph would be updated, which would then be used to update any affected technical documents.
I'm not sure how feasible this idea is, but it seems feasible, which is why I reached out. I can see how such functionality would be very useful in many applications. I'd be interested to hear your feedback.
Can you add your own apis to connect to wolfram to research equations?
No, not possible. But how would you research equations using text network analysis?
I have found the Loading of the subscription module, is not working for me. I'm using opera one. any way of getting trial without the payment module?
Do you have other payment options like paypal?, also can you change payment to own currency AUD?
Yes, PayPal exists and it might automatically convert it to AUD if you're in Australia. I'd recommend using Chrome or Firefox.
❤
Drugs
....for the mind!
❤❤❤❤