Running locally is not necessarily open source. I am quite sure Apple is cooking something we can run locally but that is not open source. And you can run elasticsearch locally which does enough of what Google does; it’s the index size you cannot run locally cheaply. With LLMs, you don’t have that issue: for the price of a car, you can run gpt4 inference at home. I guess for that same money, you can run elasticsearch with at least a good portion of the web indexed. It’ll get cheaper fast and the distance in quality between local and remote with get smaller. It has to if these companies want to make any money instead of burning through billions of investments.