> All of those where proven harmless for LLM inference.

Could you share link(s) to those proof(s)?