AI and LLMs feel too good to be true. I bet there is no single person who knows how to touch a keyboard and has not yet come up with novel ways to use AI for whatever (s)he does; not only engineers. We keep looking for the “catch”, and one of them (beyond reliability hiccups and occasional moral considerations) is the cost. Not just the current cost for the user, but the bill that may come later, when we really need to cover the full expenses of the infrastructure that provides us with all those AI goodies.

Considering the magnitude of the investment in AI infrastructure, it is difficult (at least for me) to imagine how the current pricing models can ever pay this back. There are more AI use-cases every week, but also increased investment in new models, and eventually there is only “that much” that can be paid. In my opinion, for AI to make sustainable commercial sense, two vectors of change must occur, and AI will prevail if they get to meet in the middle somewhere:

1. Customers (private and corporate) need to pay much more for AI.
2. AI computation cost needs to go way down.

Next we’ll see where we’re at today and the changes that may happen across those two vectors.

Continue reading "Next for AI: from making addicts to improving efficiency"