[deepsquirrelnet]

Moreso than AI, ML will continue to integrate into business and improve overall productivity. I don’t worry so much about monopolies in AI, because compared to the business solutions from ML — which are often much smaller, boutique models that perform more efficiently than large, general purpose models.

I believe the soon to be realized lesson is that solving one specific problem is significantly cheaper than trying to leverage much more expensive models built to solve a great many problems.

There’s still a lot of runway left in “small” models (eg BERT) that are still being researched and augmented to solve common business problems. 10 years from now, I believe some form of the current models will become industry standards as methods for solving specific problems.

[darawk]

Yep, I totally agree, as a median forecast. But the future is always uncertain, and it is possible the things we're seeing now will hit some sort of wall, where only large companies will be able to build highly general models and everyone else will be left licensing from them. I think that's unlikely, but I just wanted to enumerate it as a possibility. And when you're constructing portfolios for long time scales, you want to think about all the likely paths things might take and try to roughly assign a weight to them and construct your portfolio accordingly.