Lançou uma IA que se recusa a escrever. E isso é o ponto (testei o Jev)
Matheus Battisti demonstrates Jev, TypeSafe's specialized decision model. He walks through early access, playground probabilities, and integrates Jev into a full-stack dashboard to triage incoming support messages, comparing performance side-by-side with DeepSeek Flash.
Original by Matheus Battisti - Hora de CodarClassificationIntermediate16 min 3 secPublished Source reviewed
Before you press play
What you’ll find in the video
Jev returns typed probability choices rather than generating natural language text or code, making it suitable for classification and triage.
Developers can integrate Jev into agentic workflows via the TypeSafe skill to execute typed categorical selections on server endpoints.
Side-by-side message triage tests showed faster response cycles for Jev compared to text models, though low-confidence decisions may require manual review fallback.
Worth knowing
Gemini-assisted video/transcript review. The author notes that Jev currently exhibits higher accuracy when prompts and input choices are formulated in English rather than Portuguese.