Almost every organization has an impressive AI demo somewhere. Far fewer have an AI system doing real work every day. Most of the difficulty, and most of the value, sits in the distance between those two states.
The AI work that pays for itself rarely looks like the keynote. It looks like an unglamorous task, done thousands of times a year, finally taken seriously. A case for choosing the dull project over the dazzling one.
Most people treat a language model like a search engine, type a few words, and get mediocre results back. The difference between that and consistently useful output is structure, not magic words.
Language models are impressive pattern machines. Whether they can reason, and how far that reasoning can be trusted, matters more to businesses than almost any other question about them.
