TypeSafe's Jev model explained: calibrated confidence, constrained outputs, 'zero hallucinations'
tech_technical · reddit · 2026-09-25
A Reddit breakdown of TypeSafe's new Jev model in 5 points:
- System 1 positioning: Jev is dubbed a "System One" model, making fast, focused decisions (the name borrows from psychology's dual-process theory).
- RLCD over RLHF: Instead of standard RLHF (which can favor answers that merely sound right), Jev uses Reinforcement Learning for Calibrated Decisions — confidence should match accuracy, so uncertain answers can be routed to human review.
- Three output formats: Choice (classification), Score (rating, e.g. customer frustration), and Noul (yes/no probability) for downstream software.
- What "zero hallucinations" means: outputs are constrained to your provided formats and options — it can't invent a fourth department, but it can still pick the wrong one. Jev guarantees an allowed answer, not a correct one.
Related event: Ex-OpenAI researcher launches Jev, a non-text "System One" decision model(12 posts)→
More from Models
- Gemini 4 Carbon matches predecessor's score but explores less, gets distracted — patience_cave · 2026-11-11
- Fable 5.2 takes MazeBench frontier, matches GPT-6 Astra puzzle-solving — patience_cave · 2026-11-11
- MazeBench: Fable 5.2 scores 15%, near-frontier at half price; Gemini 4 Carbon 11% — patience_cave · 2026-11-11
- ERNIE 5.5 Flash Preview enters LMArena Battle Mode for crowdsourced ranking — arena · 2026-11-11
- Blind tests: DeepSeek matches Opus on 95% of agentic loops at 40x lower cost — bindureddy · 2026-11-11
- Indie dev flips from $20 Pro back to $200 Gemini Ultra after two days of deep use — yihui_indie · 2026-11-11