Many here know I personally despise AI, but to take off my tech critic hat for a moment, if subscription models don't work out, we still ultimately do have a bunch of pretty powerful base models to build on. If you can push inference to the edge and run on local machines, you can make your users bear the cost.
