Talking to the brain: Using Large Language Models as Proxies to Model Brain Semantic Representation