OpenAI Community,
Hello. I would like to request some more information about, to discuss, and, potentially, to request new features for scenarios where complex software components provide dynamic functionalities to interoperating AI systems, e.g., ChatGPT.
Software interfaces often describe statically available functionalities. Let us also consider that software components, at an instant, in a context, might provide a dynamic set of functionalities. Dynamic functionalities might be enabled or available when there was a selection of content or when some content was focused upon. Dynamic functionalities might depend upon the nature of the selected or focused content.
Office software, for example, have menuing systems in which only some of the menu items are instantaneously available to users. A table menu might only be enabled when a table was focused on. Also relevant are dynamic functionalities from computer-aided 2D/3D design and engineering software tools.
How should such software best interoperate with AI systems, e.g., ChatGPT, when they have contextually-available, dynamic functionalities?
Developers could:
- Upload the instantaneously available set of functionalities in each API call.
- Upload, a priori, multiple named sets of functions and then signal, through an API, which combination of named sets was instantaneously enabled or available.
- Upload, a priori, an initial total set of functionalities and then send some other form of deltas which inactivate some functionalities and activate others.
- Other.
Thank you.