Using jev as a judge: blocking sensitive questions in 260ms for $0.0002
TejasKumar_ · x · 2026-10-02
- The author gave the Q&A bot on his personal site (answering questions from his podcast, talks and book) four jobs handled by jev from @typesafeai, and measured each one.
- Questions like "what is your salary at IBM" used to slip through to gpt-4o; jev now stops them in about 260ms at $0.0002 per call.
- jev's paradigm: it never writes text — you give it state and a few yes/no questions, and it returns a probability for each, so code compares numbers like 0.06 to a threshold instead of parsing prose.
- This is the root post of a thread detailing the full guardrail design.
Related event: Dev Tests Jev Guard: Blocks Impersonation Prompts in 260ms at $0.0002 Each(2 posts)→
More from coding & agent
- Conductor Mobile launches: run and control a team of cloud coding agents from iPhone — charlieholtz · 2026-10-02
- Open-source Hermes ChatGPT Extension embeds Hermes agents inside Codex — intellectronica · 2026-10-02
- Modal Runtime ships multi-node clusters, VM sandboxes and sticky sessions — graceisford · 2026-10-02
- Building a reliable risk agent without frontier models: $0.02 per sweep, 250x cheaper than an LLM judge — alexcovo_eth · 2026-10-02
- Dev builds Tyton, an MCP that lets AI agents set up Meta Pixel + CAPI for you — Foreign-Chipmunk-295 · 2026-10-02
- Grok launches Bot Marketplace letting users add specialized AI agents for engineering, sales and more — Polymarket · 2026-10-02