Local AI-assisted CAD without tokens could save designers a ton of money, reduce privacy risks, and allow for greater flexibility to change or swap out components.

Hi MPScmhmied,

Thanks for replying.

Regarding the cost of the ChatGPT/OpenAI/others pricing for the queries, could the high costs be due to all those tokens crawling and trawling all over the planet (and possibly causing tokens expensiveness to rise with repeated, incessant "break-in attempts" because the AI's just gotta get that data, no matter what), where the AI is (behind the scenes) being (in some cases) told to doubly and even triply check sources, driving cost even higher?

As for scaled-down locally-run LLMs not running efficiently or truthfully/factually and reliably, could that deficiency be reduced by a CAD company ensuring materials sciences, FEA, FEM, CFD, strip theory, etc are added to the local AI as a sort of judge, jury, and 3rd-party tester?

To partly answer my own question, I suppose a local user could study some A/B or A/B/C/D alternatives, then continuously make the local LLM do reasonableness and sanity checks and then sign, seal, and authenticate the results with external auditors. (Maybe, maybe, insurance companies could help curb the mad gold rush toward wasteful energy burn by compelling AI/GPT companies to lay high premiums as a hedge against sloppy data interpretation and presentation? Basically, enhanced, compulsory E&O insurance. I imagine the insurance companies feeling left out of the info gold rush party.)