Skip to content
JevDirectory.org
Evaluation & Training

Generation limitation

Jev is not trained to generate text; forcing it by chaining choices is slow and unreliable, so extraction becomes a Choice instead.

Open docs.typesafe.aiAll 117 terms
Category
Evaluation & Training
Also known as
—
Related terms
4
Directory entries
2
Docs
docs.typesafe.ai
Added
2026-09-24

Definition

The jaggedness page lists generation as a failure mode with a blunt remedy: when the answer space is bounded, turn extraction into a Choice over options rather than asking for the value itself, and if you really need text generation, there are other models for that.

The pre-parsed extraction cookbook is the constructive version: regex or a generative model proposes candidates, and Jev picks the correct span so the returned value is copied verbatim.

Tagslimitationsextraction

Related terms

Definitions that connect to this one.

JaggednessThe documented set of jev-1.13 failure modes — literal reading, counting, dates, indirection, context rot — each with a guardrail.Evaluation & TrainingPre-parsed value extractionRegex finds candidate spans, then Jev picks the requested one so code copies a verbatim value that cannot be invented.PatternsChoicePick one option from a set you define; returns the selected option, a probability for every option, and confidence.PrimitivesText-only inputJev evaluates strings, JSON objects, and arrays of text; images, audio, and video are not supported and must be pre-processed.Core Concepts

From the directory

Jev 1.13 Jaggedness
A maintained list of jev-1.13's known failure modes, literal reading, unreliable counting, dates as text, indirection, context rot, and contradictory criteria, each with a guardrail.
Practices & PatternsDocs#official#docs#evaluation
Official
Cookbook: Pre-Parsed Value Extraction
Regex finds candidate emails, phone numbers, and amounts; Jev picks the requested span so code copies a verbatim value that cannot be invented, with no free-text extraction required.
Cookbooks & DemosDocs#official#cookbook#extraction
Official

From the community

Posts from builders shipping with Jev right now.

Follow @typesafeai

Pixel art out of 256 multiple-choice questions

Aryan Singh
Aryan Singh
@just_aryansingh
·

I turned @typesafeai Jev model into a pixel art model. 256 multiple choice questions, one per pixel, plus refinement passes. These were the best results out of the bunch I tested. Code: github.com/Wizhill05/type…

Image
Image
Image
Image
Diogo Almeida
Diogo Almeida
TypeSafe AI
@CompleteSkeptic

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x

Watch on X
Reply
Read 1 reply

An Elixir port of the official SDK

jod: Zod validates data, Jev judges it

Loki: a self-improving harness with a Jev companion

Jev as an LLM guardrail inside Agentgateway

Jev reranks station suggestions in TrainLCD