This is a great piece. I often observe that this really applies at a high level in general as one of the key elements that demarcates successful / advanced econ

This is a great piece. I often observe that this really applies at a high level in general as one of the key elements that demarcates successful / advanced economies from lesser ones : how much is trust a limiting factor on what you can achieve? If you think about modern society and the levels of trust needed to do certain things : I spent nearly a lifetime of earned income on a house and I completely trusted the agent and bank who facilitated that. Without that trust, the whole transaction wouldn't be possible. Why did I have that trust? These are the hidden value of all those laws and regulations that we often hate so much. As much as anything they create the stable platform for high value transactions to take place. The price of not having them is not being able to access the value you can get if you can successfully execute high trust transactions. There's no way I would gamble my whole life's income if I didn't trust the process. I would have to buy a smaller house. So AI is playing out along these lines. How much of the value we can access from AI is eventually going to be limited by how much trust we can grant it. Which ironically may derive from how much regulation / legal framework is successfully put around it, which may in the short term slow it down. The self-hosted open source model route is a substitute / fallback for not having trust in hosted models. We can't really pretend that it's possible to self-host a model that will be as good as a hosted one, the scale is always going to be a material difference. But we may be to hit the point where for targeted applications there are diminishing returns, and that may be good enough.