I read this paper (https://arxiv.org/pdf/2410.13722) on pre-training poisoning of LLMs. The authors demonstrated that poisoning 0.1% of the pretraining dataset can measurably manipulate beliefs of the model.
What I read about is a more sophisticated supply chain attack. But these days, search previews use retrieval-augmented generation to parse websites to generate answers on the fly. In that case you don't need to poison the supply chain; good SEO is enough.
Congrats! Happy for your success. Your mix of ingredients is impressive (the hardware alone is probably simple). Your product idea came from actual personal need (validation!), and you are able to tie the hardware to your software to widen the moat -- which is great.
Your anti-counterfeiting tech hopefully / probably prevents your Chinese manufacturers / suppliers / related entities from free-riding on your work. (If your volumes start hitting say 50K+, things may change, you may be entering dangerous territory then, so I'd suggest you spend some time on beefing up the moat).
Yes, that's the usual approach. Mylar sheets for short runs, stainless steel sheets for production.
3D printing in PLA probably won't have enough dimensional accuracy for fine pitch boards or boards of any size. The demo shown is a very small coarse pitch board.
The thickness of the stencil matters, because that determines how much solder paste you are laying down. This approach is going to produce thick stencils.
There are CNC machines for laying down solder paste. Here's one.[1] It does a few more jobs, too. The various CNC desktop board-making machines are still too expensive for most hobbyists, though.
You could solve the inaccuracy by just using a resin printer, that should have enough accuracy (maybe with some fiddling with the settings/trying out resins, light can bleed to the sides of the tiny holes needed). Although I have never done a stencil that was very fine but did fine silicone moulds with them.
