I think there’s a really interesting point here that even if a model is capable of planning and reasoning, part of the skillset of a creator is that asking for the right thing requires an understanding of how that model is creating its artifacts.

And in 3D, you won’t be able to do that without understanding, as an operator, what you’d want to ask for. Do you want this specific part to be made parametrically in a specific way for future flexibility, animation, or rendering? When? Why? And do these understandings of techniques give you creative ideas?

A model trained solely on the visual outcome won’t add these constraints unless you know to ask for them.

Even if future iterations of this technology become more advanced and can generate complex models, you need to develop a skillset to be able to gauge and plan around how they fit into your larger vision. And that skillset requires fundamentals.

The article is about language models. Language models are not capable of planning or reasoning.