JevMade hello@JevMade.com
← Back to agent workflows videos

JevMade field notes / Video guide

【徹底解説】「ハーネスの常識が変わる」と話題の“判定専用AI”『Jev』を使いまくった結果、確かに強力でした

This video breaks down TypeSafe AI's Jev model, explaining its typed decision structure, practical optimization tips for states and criteria, hands-on empirical comparisons against LLMs, and how to effectively divide labor between code, Jev, and generative LLMs.

Original by まさおAIじっくり解説chAgent workflowsIntermediate28 min 37 sec Published Source reviewed

Before you press play

What you’ll find in the video

  1. Jev evaluates supplied state against bounded question types and returns probabilities or scores rather than prose.
  2. The author recommends concise state, observable criteria, and task-specific thresholds after seeing misclassifications in exploratory tests.
  3. The proposed harness uses Jev as an intermediate evaluator while keeping exact operations in code and harder reasoning with a generative model.
Worth knowing

Gemini-assisted video/transcript review. Speed and low cost do not guarantee decision accuracy; author experiments showed Jev misclassified outlier low-performing content and required prompt tuning to produce reliable scores.

【徹底解説】「ハーネスの常識が変わる」と話題の“判定専用AI”『Jev』を使いまくった結果、確かに強力でした