No signs of AI payoff outside the tech sector

We at Reliath AI see something similar. There are many conflicting demands on companies. They think that they need AI to be competitive, but they don't have the knowledge to translate that into specific actions. What do they want to use AI to do? By default it simplifies into writing summaries. As AI is used more and more to write, the bottleneck to communication moves from writing to reading. The volume of AI writing overwhelms the ability of users to read it. So they get their AI to summarize the writing that another AI expanded from a summary. It reminds of the scene from the movie Real Genius where the students recorded a lecture presented by the professor's cassette player.

The cost of individual tokens has dropped, but the number of tokens needed for a task has exploded. One legal company we know had their token spend increase 13x from January to May this year. Their fix was to move to a self-hosted model. They cut their token spend, but increased their infrastructure spend.

At Reliath AI we do not bill for tokens or for usage. The Reliath AI models are built on facts, not tokens. We use a tiny fraction of the computational resources of the Large Language Models because we are not building large language models, but bespoke fact models. We charge by how much the model knows, not how many tokens it can burn.