If there is anything I can expect to remain consistent over the next 30-40 years, it's that the majority of people have no interest or ability to maintain technical systems themselves, as evidenced by the often banal and extremely simplistic tasks that are required of me as an IT support technician, family, friends, etc. There absolutely will be a direction in models toward more specific tasks, but they will be packaged and distributed by channels and entities that need to monetize their services in some way. That is just inevitable. The problem was never the access to information, it was the nature of most people. (Just to be clear, I don't think this is technically a bad thing, just something I have noticed.)

> it's that the majority of people have no interest or ability to maintain technical systems themselves

In the same way you likely don't grow your own food or make your own shoes.

I think it'll mostly still be centralized around providers

I don't see good local models happening on mobile devices anytime soon, and the majority of desktop users will use whatever is the best + most convenient. Competitive open source models running on your average laptop? Seems unlikely

Doesn’t seem that unlikely to me. Ollama on Mac can already run decent DeepSeek/Llama distillations. For a lot of tasks it’s already good enough.

And 2029 is in 4 years. Four years ago leetcode benchmarks still meant something and OpenAI was telling us GPT3 was too dangerous to release.

Remember when kids used to learn editing or photoshop? You can do that in some remarkably capable tools online, and the kids don't bother with offline tools anymore. This is the same thing.