> So maybe this technique will work when we all have our own locally hosted AI?
None of the trillion-dollar corporations working to make AI the next big thing want us to have locally hosted (and certainly not locally-controlled) AI.
The trillion-dollar corporations are leading the way, but it's doubtful that they will get to control what sort of AI gets developed over the longer term. I expect that corporations with only tens of billions of dollars, and even universities or charities with single-digit-billions endowments, will eventually be able to develop models that would be considered state-of-the-art as of 2024.
Most of these smaller organizations don't have the same incentives to preserve advertising that the trillion-dollar corporations do. If Microsoft, Meta, or Alphabet think that they're going to dictate AI 10 uses years hence, they're as over-optimistic as those Americans who hopefully predicted in 1945 that the Soviet Union might take another 15 years to build their own atomic bombs. The USSR actually tested their first atomic bomb in 1949.
>The trillion-dollar corporations are leading the way, but it's doubtful that they will get to control what sort of AI gets developed over the longer term
Google and Apple have replaced the concept of a 'computer' with dumb clients called 'phones' and 'tablets' in less than 20 years. The corporations have already won - B2C AI will be closer to Clippy than offer any of the high-minded use cases discussed here.
Whatever open-source, on-premise AI available to the individual is going to be a hard-to-use, Linux-only mess that will scare away the average person.
None of the trillion-dollar corporations working to make AI the next big thing want us to have locally hosted (and certainly not locally-controlled) AI.