Is there something like this planned for Rhino 10? I guess it may not be optimal for McNeel to train its own model from the ground up, but could Rhino adopt a framework so that open models can interact with it? Like it’s own “Hugging Face”. There is alreay a jewelry plugin (Rhino Artisan) using basic AI to create specialized jewelry componentes directly in Rhino.
The ideal version for me would be a model running locally, with the hability to ingest my library of designs, and learn from that, beyond basic modeling capabilities.
Given all evidence is that LLM’s don’t actually improve productivity for programming, the most obvious use case in the world with the mountain of training data available, this is likely to amount to a pile of nothing for the foreseeable future.
We used GPT for all the Arduino programming on a motion control system. 8 motors, sensors etc. all coordinated and working 100%. Granted it’s not that complex, but in a similar vein there’s people who don’t use CAD who would use AI tools to generate parts.
AI might not be great at doing what you do, but it’s decent at opening doors to doing things you wouldn’t usually do.
Your anecdotes mean nothing. The actual research shows they make productivity WORSE(while people think it makes them better)and there is no tsunami of silly little apps that we would see if it really did work.
Interestingly, I had different systems fail to write working code for a seemingly banal impact sensor threshold problem (trigger an LED when a certain threshold is crossed, but only the LED associated with that threshold, and not those associated with lower thresholds). It were the innards for an interactive martial-arts “Wing Chun training pad”. AI did not figure out that one needs to sample discrete time intervals with a self-tuning interval duration and came up with silly state-machine code suggestions that did not work.
