demonstrate and guide chatGPT driven process editing

[ivy-rew]

• highly reproducible examples to edit existing processes
  • nice template for 'inserting' new elements
  • hints for chat-gpt; use the schema or something the like
• inform @andreasbalsiger

[ivy-rew]

just invested a few minutes to find ways to make chatGPT supply more reliable results when being instructed to add new elements onto existing processes. In short, there is no fast and easy solution :)

• we can improve the outcome by shipping constraints, that should always be respected. See this example: https://www.reddit.com/r/ChatGPT/comments/zlo9yv/json_as_prompts/
  • Now we can make use of this:
    • to get valid elementIds, we just have to tell chatGPT to make it as designed.
    • or to ensure that new elements will automatically be connected or inserted at the end of the existing landscape.
  • We could such a more complex request to chatGPT without making the users know. Its just to supply as much context that we already know as we can. taskrules.json.txt
• new elements are added appropriately configured, if there are already other instances of such an element within the process. so maybe we should bend our promoted use-cases onto refactoring tasks, rather than adding completely new stuff onto virgin processes.
• json-schema: unfortunately I could find no rock solid way to let chat-gpt interprete the json-schema which would help the AI to find all valid combinations. This is really cumbersome but maybe I just didn't try hard enough. However, basically anyone willing to help here can do it by using the playground. I just added an empty json.process of us into it and experimented with the different engines and its settings. https://platform.openai.com/playground?mode=complete&model=text-davinci-003
• another way to proceed could be to submit concrete training data; which seems cumbersome to do, but would definitively help a lot for everyday AI requests. See https://platform.openai.com/docs/guides/fine-tuning to dive into that topic.

[ivy-rew]

documenting past attempts to make chat GPT comply with our processes:

the naive approach

1. use chat-gpt context-menu, ask to insert an element (hinting the json array name and the 'type')
2. accept the incoming change
3. now we need to "fix" chat GPTs answer with a real editor, that has basic json-schema support.
4. the process is back functional after this.

[ivy-rew]

looks as if openai finally support schema based outputs, definitively what were looking for in the past
https://openai.com/index/introducing-structured-outputs-in-the-api/?ck_subscriber_id=2712594844
maybe you want to plan another exploration story in this field @ivy-sgi ?

[ivy-sgi]

Thanks for the hint @ivy-rew We will have a deeper look into this.