The mainframe model fell apart the moment that microcomputers became powerful enough to satisfy same use cases sufficiently. Centralized GenAI will also become obsolete as soon as local LLMs are capable enough to satisfy the same use cases sufficiently.
Artificial lock-in simply doesn't work in the long run: the incentive structures will always motivate customers to cut out middlemen, and peripheral markets to develop around providing the tools for doing exactly that. Anthropic and OpenAI may well end up being the Data General and Honeywell of our era.
The greatest risk to this is the possibility of political intervention creating artificial hurdles that prevents decentralized AI from challenging the big players. With than in mind, it's worthwhile to subject every proposal to regulate AI to intense scrutiny.
This calls to mind the war on general-purpose computing (https://boingboing.net/2012/01/10/lockdown.html) and it amazes me how even today we are still stuck with a couple of companies that have already cornered their markets, and yet still won't give up their fight to take microchip-technology out of the hands of their fellow humans - still trying to move the whole part of executing commands back within their own walls, and have people subscribe to have access to being able to request a specific type of process to be applied to their input. It occurs to me that surely all this must be the result of some ego/power-trip, for it hardly serves any party in any future I can envision where the ability to have computers compute is placed under lock and key, out of reach of the general public.
Is it simply a couple of billionaires eager to pull tricks like Adobe did when they cut an entire country off from access and use of Adobe-software, just for the thrill of it? Or is there actually some plausible future benefit or a specific outcome they have in their minds, and am I (or are we) too ignorant to be able to see anything worthwhile in their direction?
Why allow the sale of personal computing-devices in the first place, if you don't want people to decide which instructions they want to feed to it's processor? Right now they may be slowing down many processes, both computational and mental, wasting lots of time and making everybody hate subscription-models more and more every minute... what is it they really hope to accomplish, apart from pissing everybody off?
