Jev, built by an RLHF/InstructGPT veteran, targets machine decisions with new RLCD training
MaryamMiradi · x · 2026-09-21
TypeSafe, founded by Diogo Almeida (who worked on the RLHF/InstructGPT lineage behind ChatGPT), released Jev — a model built for software to make decisions with, not for humans to chat with. Its RLCD (Reinforcement Learning for Calibrated Decisions) method optimizes for calibrated machine-consumable decisions instead of human-preferred responses. A 6-step explainer thread.
More from Models
- Anthropic teased to have big news next week, as 'slow down' talk fades — iruletheworldmo · 2026-09-21
- GLM-5.3 FlashX Now Available via Nous Portal and OpenRouter — Teknium · 2026-09-21
- Tuned 26B diffusion model beats Jev after adding 512-token decode-time reasoning budget — rickasaurus · 2026-09-21
- Users Report Opus 5 Feeling Suspiciously Faster and Better Inside Claude Code — rudrank · 2026-09-21
- Hemmingway-1: Apache-2.0 27B writing-only model scores 1330 on EQ-Bench 4, beats GPT-5.5 — Lukinator6446 · 2026-09-21
- Astra is terrified of mistakes — humans may still own high-entropy decisions — xiaosun86 · 2026-09-21