The point about the economics of running these models is an important one that slides under the radar a lot of times. The training costs for large language models like GPT are enormous, and the inference costs are substantial too. Right now things like ChatGPT are very cool parlor tricks, but there's absolutely no way to justify them in terms of the economics of running the service today.

Obviously this is all going to change in the near to mid future: innovation will drive down costs of both training and inference, and the models will be monetized in ways that bring in more revenue. But I don't think the long term economics are obvious to anyone, including Google or OpenAI. It's really hard to predict how much more efficient we'll get at training/serving these models as most of the gains there are going to come from improved model architectures, and it's very difficult to predict how much room for improvement there is there. Google (and Microsoft, Yandex, Baidu, etc.) know how to index the web and serve search queries to users at an extremely low cost per query that can be compensated by ads that make fractions of a cent per impression. It's not obvious at all if that's possible with LLMs, or if it possible, what the timescale is to get to a place where the economics make sense and the service actually makes money.