[davidpolberger]

Does anyone know if this works well (to the extent that it does), because the documented Blender Python API was part of the GPT-4 training set, or because the Blender API is fairly predictable?

I'd love to add this capability to our SaaS product, but I've waited for OpenAI to make GPT-3.5 or GPT-4 available for fine-tuning. (Cramming an entire API into the prompt does not seem feasible, not even with support for 32K tokens.)

[iamflimflam1]

He's using the fact the the blender python API is part of the training set. The prompt used can be seen in this file: https://github.com/gd3kr/BlenderGPT/blob/main/__init__.py

[VadimPR]

I see the prompt but not the part where he's supplying the API - mind pointing me to it?

[MattRix]

He doesn’t need to supply the API, GPT already knows the API from its training data. I’ve written a similar tool for Unity and it works the same way. Any program that allows scripting/plugins (and has a sizeable community posting code online) will be able to work the same way.