counter example: me! autocorrect, spam filters, search engines, blurred backgrounds, medical image processing, even revenue forecasting with logistic regression are “AI” to me and others in the industry

I started my career in AI, and it certainly didn’t mean LLMs then. some people were doing AI decades ago

I would like to understand where this moral line gets drawn — neural networks that output text? that specifically use the transformer architecture? over some size?

You take a pile of input data, use a bunch of code on it to create a model, which is generally a black box, and then run queries against that black box. No human really wrote the model. ML has been in use for decades, in various places. Google Translate was an "early" convert. Credit card fraud models as well.

The industry joke is: What do you call AI that works? Machine Learning.