AI multiple choice, not essay writing

Yes, and the post says so: the figures come from TypeSafe's own workflow benchmarks. A framing thread is not a place to look for independent numbers and the replies do not add any. The part worth keeping is directional, a fraction of a second per call against multi-second frontier calls, rather than the exact multiple.

You keep both. The post's advice is to hold the frontier model for hard thinking and writing and use Jev for the rapid-fire decisions in between, with examples that are all control-flow questions your code can act on: spam or not, which tool this agent should use, does this need a human. That honesty about scope is also why it cannot help with generation at all.

It holds for the shape of the output, not for the reasoning. A typed question with fixed options and a probability per option really is closer to classification than generation, and dropping that into ordinary control flow is the useful part. What the framing hides is how the judgment is reached, and the post does not say how choices are made or how to check them.

One reply claims it, unverified. A builder says an agent he built to run a hotel switched from a frontier model for classification to Jev, calls it substantially better, and matched workflow routing 100% on things like guest requests and lockouts. It is a single unreplicated claim with no method or numbers behind it.

Matt Van Horn, who says he co-founded June, a self-driving oven company acquired by Weber, and the company that became Lyft, and who maintains a project called slashlast30days. So the explainer comes from a consumer-product founder reading Jev as a builder rather than from a model researcher.

Yes, he turned it into an article titled WTF Is Jev? 9 Things People Are Already Building With It. The article keeps the same multiple-choice framing and adds a list of launch-week uses, so it is the fuller version of this post rather than a separate source.

That it was made by Diogo Almeida, whom he calls the co-creator of ChatGPT. He drops this in a follow-up reply as an aside rather than as evidence, which fits the thread's tone: the multiple-choice framing does the explaining and the pedigree is a bonus rather than the argument.