No, your premise is false. AI is neither dangerous nor a weapon. Therefore the rest of your statement is invalid. Try again with a new premise that isn't false.

dual-3090 + sli - 48gb vram - ~$3000 (7b-34b inference, maybe tiny finetuning) quad-mi100 + infinity fabric - 128gb VRAM, ~$6000 (good for finetuning and inference) m3 max 128gb - bad for finetuning, great for single-user inference. - ~$6000

I trained it with no refusals in the data set But the base model still has it's opinions You have to use system prompt. https://github.com/ehartford/dolphin-system-messages

I'll put the 4090s on that one, they have video output

Ok I'll make a blog post

On windows, this is tested in WSL2 not in native. You might be able to get it to work in native windows, with enough tweaking

is it possible using faiss instead of pinecone?

haha, gotta leave something as an exercise for the reader