@hongminhee The current crop of LLM cannot be claimed for the public good. They were built using an insane amount of hoarded capital, and are constantly trained on the entirety of humanity's intellectual property without asking for consent. If you come short in either aspect during a fictional reclamation process, you will have a substandard LLM system.
Besides, the entire point of "reclamation" assumes the current LLM systems are well-established, when they are only a few years old and fledgling. All the major models are currently heavily subsidized, will you still using the same translation service when it starts charging you 2, 5, 10 or 50 times more?
Uber started the same way, by offering cheap and convenient taxi rides to users, and a cheap path to a medallion-like situation for drivers. It succeeded at destroying the taxi ecosystem which, for all its flaws, was ensuring a real livelihood to a limited number of drivers. Now there can be an unlimited number of drivers earning precarious wages while Uber acts as a toll both for every single ride.
Similarly, the path to profitability for the current "AI" companies depends on destroying an entire class of intellectual professions and inserting themselves as toll booths between people and the entirety of humanity's intellectual works, indefinitely. There is no reclamation here, no possible ethical local models, it's an all or nothing situation.
But it isn't too late. There's no profitability yet, which means there can be a "market readjustment" (=financial crash) which will likely wipe the largest and most expensive models to train. The ones that will remain will be more modest, less generalist, more specialized, just like before ChatGPT.
And yeah, machine English-Korean translation may suffer in the process and I'm genuinely sorry about that, but the issue is so much larger than just this particular use case.
