Wanted to suggest HRM model for creating Geonodes setups

Geonodes is complex. I haven't played around with it yet using claude and such, but I imagine that LLMs might have a hard time with more complex setups, starting with a pre-trained one and fine-tuning perhaps.

I came across the Hierarchical Reasoning Model stuff lately and thought that it might be a really good fit for a NN that creates geonodes setups.

For loss calculations, you could use an Agent-as-a-judge technique whereby the agent runs and tests the setup for prompt adherence and whatever else is defined in your rubric. All funded by NVIDIA.

It wouldn't be easy - a project - bringing in lots of geonodes setups. Perhaps the community can all upload theirs to a webpage to grant rights to use as training data - everyone wants to see Blender progress, so I think the community would love to donate for the training. They don't have to give their stuff away, but just for contributing to training efforts.

Need AI as a new category for RCS.