Choice, Score and Noul: four mistakes with Jev's primitives
Jev is a System One model from TypeSafe. It reads natural language like any LLM, and instead of writing a reply it returns a probability distribution over options you supply. No prose, no reasoning trace, no JSON to repair — a number per option, summing to 1. I spent a week building a Magic 8 Ball on it, which sounds like a toy and turned out to be a decent test rig: every answer is a judgment…
I constructed a Magic 8 Ball using System One model Jev from TypeSafe. Jev reads natural language and returns a probability distribution over the options you provide, with no prose, reasoning trace, or JSON repair. I made four mistakes in building my Magic 8 Ball, each costing me about $0.000014. The first mistake was interpreting the Score as a winner.
The Score returns a position on your levels, but it's actually a probability-weighted average. The second mistake was assuming low confidence indicated doubt about the answer. Every Choice and Score comes with a confidence rating from 0 to 1, but confidence measures how peaked the distribution is, not the truth of the answer. The third mistake was offering options that split their own vote.
An 8 Ball typically has twenty answers, and my ball only had five. When I asked my ball a question, it offered multiple answers that were near-synonyms, causing a disagreement in the options presented. Finally, I forgot that the options are part of the prompt. The model cannot choose an omitted value; leaving an option out makes it impossible for the model to provide that answer.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
