A recently leaked internal document from Google indicates that the market share of big tech for large language models is not as secure as it seems thanks to advances in open-source LLMs.