The Jev and “system one” models look interesting. Being able to consume the model results in a structured way would simplify the design of services that could use LLMs. Post category classifiers, for example: “does this post feature walking outside or travel?” If yes, set category to “Outside”.