This video breaks down TypeSafe AI's Jev 'System 1' model, demonstrating its Choice, Score, and Noul primitives through a custom Japanese demo app. It explores latency, pricing, prompt injection vulnerabilities, multi-query batching, and what 'zero hallucination' strictly means.
Original by 中村祐太のFindUアカデミー【AI&WEB開発】ClassificationIntermediate20 min 28 secPublished Source reviewed
Before you press play
What you’ll find in the video
The Japanese demo app asks Choice, Score, and Noul questions and displays their probability-bearing answers.
The presenter’s adversarial wording changed classification probabilities, showing why a fixed answer schema does not establish prompt-injection resistance.
The batching demonstration asks several questions over one state; its timing is a local observation, and a valid output shape does not prove a correct answer.
Worth knowing
Gemini-assisted video/transcript review. Demonstration figures reflect exploratory single-environment runs rather than controlled benchmarks; non-English performance and prompt injection resistance require rigorous task-specific validation.