Testing Jev, an RLCD decision model, to build topical maps of Anthropic's 547-page blog
cneuralnetwork · x · 2026-10-21
The author tested Jev classification across Anthropic's 547-page blog inventory with striking results. Jev is a System One model built on RLCD (Reinforcement Learning for Calibrated Decisions): it acts as a decision layer, returning a choice/score/noul value per question, keeps contexts isolated between questions (where RLHF models leak), charges nothing for output tokens, and is very fast.
Workflow: code fetches a site's sitemap, an LLM names categories/subcategories, and Jev maps every page to topics — producing a topical map of Anthropic's site (AI Safety Research, Claude Models, etc.) and flagging 27 pages with no topic.
More from coding & agent
- Production Debugging Playbook: a six-step method to find the real cause of 5xx spikes — _jaydeepkarale · 2026-10-21
- Browser Use open-sources rl-env-use: one-line RL environments for any domain — inductionheads · 2026-10-21
- Uber's AI System AutoClean Deleted 1M+ Lines of Dead Code Across 16,000 Stale Feature Flags — blaizedsouza · 2026-10-21
- Turning fluid intelligence into crystallized: why agents should cache repeated solutions — AdaptiveAgents · 2026-10-21
- Who Pays When Actions Call Actions? The Missing Execution Model for Internet-Scale Agents — AdaptiveAgents · 2026-10-21
- Turn Repeated Agent Compositions into New Actions: Fluid to Crystallized Intelligence — AdaptiveAgents · 2026-10-21