Indeed you have cracked the code for BPMN XML, thanks for what you've demoed. I'll have to give it a try as I'm interested in how events, both interrupting and non-interrupted can be modeled with ChatGPT. Thanks again.
I found that Sonnet 3.5 excels in creating diagrams, whether in Mermaid format or others. However, achieving the best results relies heavily on using effective prompts.
Yes, you’re correct. All additional model actions like vision, webaccess, text input are still disabled. I believe they will be added later though. You can also switch between the models per question, meaning that technically it could be possible to use GPT 4 for the scrape and then o1 for the continuation. Nice catch!
Nice coverage. I actually found sonnet 3.5 was really good with diagrams whether mermaid or others. But the only way to get it right is using really good prompting.
My issue with Sonnet is the type of diagram generated,...and the layout of the black background is terrible. So I usually download the Mermaid file and generate it again on a Mermaid Reader...that quality there is better
Thanks so much for posting this video. Have you been able to take the code /diagram and put it through Viseo or Lucidchart to create a colouredBPMN diagram?
@@Gemma-v2l you’re very welcome! I haven’t tried that yet :)! Let me know if you get that to work. Stay tuned for updates though, I’m building my own BPMN AI tool that high likely will have colored elements!
this is not really working. I followed exactly your prompt/instructions (to the letter and ponctuation) and still encounter several problems. The biggest issue is your prompt at 4:45. I ran the same prompt several times, sometimes it uses placeholders, sometimes not. In general, the content of the XML show many differences after each attempt. my first 2 attempts, the XML was totally rubbish, your website would not accept it and generate this error: "Error rendering diagram: no diagram to display" on my 3rd attempt (again: exact same prompt), it finally generated a XML that didn't cause an error on your website but all connectors were missing. After analysis of the XML structure, I notice there were no connectors. WTF? I told chatGPT that the connectors are missing and it just says: "sorry for the oversight, here is the corrected version..." and indeed, connector were present in XML and your website would show them. This just illustrates how poor chatGPT really is. It's fundamentally unreliable as repeating the same inputs don't produce the same output. when you work with code, XML, you can quite easily see that something is wrong, at the very least when code causes an error. When you work with generated text, it's much harder and time consuming to detect the bullshit (it often quotes sentences from people don't even exist and then it just apologies for the oversight...)
Well I do have to say that I agree on some of your points there. And unfortunately when it comes to “generative” AI, hallucinations are probably there to stay as they are fundamentally linked to the way the transformer models are working. In terms of the prompt in the video, I have since then updated the prompt and included more of the BPMN diagram language in it. It provides a higher successrate that way. But at the same time I’ve realized that there are way better ways to achieve this outcome. I’m currently building a BPMN genAI tool with a few friends that I’ll showcase here when it’s ready. We are utilizing GenAI for the bits that are relevant and stick to other methods when it comes to getting successful results. I’ll make sure to send you a message so you can see if it works better when it’s ready ✌🏻 thanks for being here and taking the time to comment under the video for others to learn 😁
what frontend do you use? I use the openAI website but it behaves differently: when I entered your first prompt at 3:04 (using o1-preview) it produces a block of text, without any formatting: Process Name: Grant Application Process Start Event: "Application Received" Task: "Review Application" Gateway: "Is Application Complete?" If Yes: Task: "Assess Eligibility" Gateway: "Is Applicant Eligible?" If Yes: Task: "Approve Grant" Task: "Notify Applicant of Approval" End Event: "Grant Approved" If No: Task: "Notify Applicant of Rejection" End Event: "Application Rejected" If No: Task: "Request Missing Information" Task: "Await Applicant Response" End Event: "Application Pending"
@@TheArtificialAnalyst A typical example might be a CRM workflow that utilises a number of linked tables ie. companies, contacts, opportunities, and approved ('live') projects. Companies may link to multiple contacts and multiple opportunities; opportunities could link to approved projects. The data set for each table could be fairly 'standard'. The objective of the diagram is to represent the relationship between the tables, the specific index for each table, and the linked fields. Ideally, annotating the field types is also useful. I'm sure that you've worked with similar diagrams previously. Hopefully this makes sense. :)
Hey there! Great question! BPMN, or Business Process Model and Notation, is all about visually representing processes in a way that everyone can understand. It helps organizations streamline operations and improve efficiency. As for automation, absolutely, but processes hide in many different corners. Think of systems, peoples minds, and sometimes processes just exist without anyone knowing. So we probably cannot automate it for the whole world (imagine the chaos!), we can certainly automate many processes within individual organizations to save time and boost productivity. Thanks for tuning in my friend!
This is awesome, nice work Pascal! I love seeing awesome examples like this leveraging the power of o1 to do something you couldn't beforehand!
Indeed you have cracked the code for BPMN XML, thanks for what you've demoed. I'll have to give it a try as I'm interested in how events, both interrupting and non-interrupted can be modeled with ChatGPT. Thanks again.
I found that Sonnet 3.5 excels in creating diagrams, whether in Mermaid format or others. However, achieving the best results relies heavily on using effective prompts.
Oh, maybe the o1-model has currently no access to the web? It may recalled how BPMN works just from its memory? [4:07]
Yes, you’re correct. All additional model actions like vision, webaccess, text input are still disabled. I believe they will be added later though.
You can also switch between the models per question, meaning that technically it could be possible to use GPT 4 for the scrape and then o1 for the continuation.
Nice catch!
Nice coverage. I actually found sonnet 3.5 was really good with diagrams whether mermaid or others. But the only way to get it right is using really good prompting.
Oh awesome! Which other languages are you using? PlantUML, Mermaid, any others:)?
My issue with Sonnet is the type of diagram generated,...and the layout of the black background is terrible. So I usually download the Mermaid file and generate it again on a Mermaid Reader...that quality there is better
This is great, thx for this!!!
Thanks so much for posting this video. Have you been able to take the code /diagram and put it through Viseo or Lucidchart to create a colouredBPMN diagram?
@@Gemma-v2l you’re very welcome!
I haven’t tried that yet :)!
Let me know if you get that to work. Stay tuned for updates though, I’m building my own BPMN AI tool that high likely will have colored elements!
this is not really working. I followed exactly your prompt/instructions (to the letter and ponctuation) and still encounter several problems.
The biggest issue is your prompt at 4:45.
I ran the same prompt several times, sometimes it uses placeholders, sometimes not. In general, the content of the XML show many differences after each attempt. my first 2 attempts, the XML was totally rubbish, your website would not accept it and generate this error: "Error rendering diagram: no diagram to display"
on my 3rd attempt (again: exact same prompt), it finally generated a XML that didn't cause an error on your website but all connectors were missing. After analysis of the XML structure, I notice there were no connectors. WTF?
I told chatGPT that the connectors are missing and it just says:
"sorry for the oversight, here is the corrected version..." and indeed, connector were present in XML and your website would show them.
This just illustrates how poor chatGPT really is. It's fundamentally unreliable as repeating the same inputs don't produce the same output. when you work with code, XML, you can quite easily see that something is wrong, at the very least when code causes an error. When you work with generated text, it's much harder and time consuming to detect the bullshit (it often quotes sentences from people don't even exist and then it just apologies for the oversight...)
Well I do have to say that I agree on some of your points there. And unfortunately when it comes to “generative” AI, hallucinations are probably there to stay as they are fundamentally linked to the way the transformer models are working.
In terms of the prompt in the video, I have since then updated the prompt and included more of the BPMN diagram language in it. It provides a higher successrate that way.
But at the same time I’ve realized that there are way better ways to achieve this outcome.
I’m currently building a BPMN genAI tool with a few friends that I’ll showcase here when it’s ready. We are utilizing GenAI for the bits that are relevant and stick to other methods when it comes to getting successful results. I’ll make sure to send you a message so you can see if it works better when it’s ready ✌🏻
thanks for being here and taking the time to comment under the video for others to learn 😁
what frontend do you use? I use the openAI website but it behaves differently: when I entered your first prompt at 3:04 (using o1-preview) it produces a block of text, without any formatting:
Process Name: Grant Application Process Start Event: "Application Received" Task: "Review Application" Gateway: "Is Application Complete?" If Yes: Task: "Assess Eligibility" Gateway: "Is Applicant Eligible?" If Yes: Task: "Approve Grant" Task: "Notify Applicant of Approval" End Event: "Grant Approved" If No: Task: "Notify Applicant of Rejection" End Event: "Application Rejected" If No: Task: "Request Missing Information" Task: "Await Applicant Response" End Event: "Application Pending"
Got my sub. ✊
This is excellent!
Now if we can just get AI generated entity relationship diagrams...
I think I can help you with that:) just comment a use case or an example that you would like to see and I’ll take a look!
@@TheArtificialAnalyst A typical example might be a CRM workflow that utilises a number of linked tables ie. companies, contacts, opportunities, and approved ('live') projects.
Companies may link to multiple contacts and multiple opportunities; opportunities could link to approved projects.
The data set for each table could be fairly 'standard'.
The objective of the diagram is to represent the relationship between the tables, the specific index for each table, and the linked fields. Ideally, annotating the field types is also useful.
I'm sure that you've worked with similar diagrams previously. Hopefully this makes sense. :)
I dont have gpt 4 version .. So didn't work this tool
What is bpmn for, why do they use it?
Can you automate it and do all the bpmns for the whole world? Lol
Hey there! Great question! BPMN, or Business Process Model and Notation, is all about visually representing processes in a way that everyone can understand. It helps organizations streamline operations and improve efficiency.
As for automation, absolutely, but processes hide in many different corners. Think of systems, peoples minds, and sometimes processes just exist without anyone knowing. So we probably cannot automate it for the whole world (imagine the chaos!), we can certainly automate many processes within individual organizations to save time and boost productivity.
Thanks for tuning in my friend!
@@TheArtificialAnalyst thanks for the speedy reply and good explanation, much appreciated.