Ask HN: What would it take to provide free AI to the underprivileged?

The good news is the tools are already there. Models like Llama, Mistral, and Qwen are free and open source. Good enough for real tasks like answering health questions, helping students learn, or explaining legal rights.

The hard part is not the AI. It is running it. You need powerful computers called GPUs to make it work at scale. Renting them from the cloud costs real money. The smarter path for a government project is to partner with local universities or phone companies that already have the hardware and buildings to run it.

The other big problem is relevance. Most AI models learn from English websites. They give poor answers about local laws, local farming, and local medicine. The real investment needs to go into teaching the model your local languages and local knowledge. That is what makes it actually useful to people on the ground.

What languages and use cases are you starting with? That changes everything about the technical approach.

if there is some way to decentralize the compute properly, it might be possible to envision some kind of AI that you can use if you allocate it some compute to use or something. like sharing ur device's resources or so on the network to get some perks on it like more resource provisioned is more context window.

the problem is that if its centralized like now, it costs a huge amount for the centralized party to run the operation. ofc not as much as they earn, so it could likely be much cheaper but it will cost electricity, hardware etc. etc.

now, i dont know how to 'decentralize' and i know its kinda a buzzword. but in my mind it would only work.

bundle all ppls resources to be a giant super computer. distribute queries over nodes that are not busy with normal user behaviours etc