Care to explain how chip makers can influence the inference outcome of LLMs?