Jev, a pick-an-option model with free answers: promising for logs and fines, but no explanations
KrstABot · reddit · 2026-09-26
The author breaks down Jev, a model by typesafe ai released September 15: it generates no text — you give it a question and answer options, it returns its pick plus per-option probabilities. No training needed, input costs $0.042/1M tokens, answers are free.
Two candidate use cases, with caveats:
- Grafana logs: huge volume, short decisions (page on-call / open ticket / ignore) — a natural fit, but Jev's own docs warn accuracy drops with irrelevant input, and logs are mostly noise; feeding pre-computed alerts may work better.
- Government fines: high-volume, homogeneous decisions suit digitally advanced countries (Korea, Japan, Singapore); code checks dates and amounts (a documented Jev weakness) while Jev handles only the fuzzy matching.
Key risk: errors come with only a probability, no explanation — a 0.95-confidence mistake would be nearly undetectable. The author suggests automating only obvious "no fine" cases, with humans confirming actual fines; where to set the confidence bar remains open.
More from Models
- Baidu says RSI auto-optimized ERNIE inference, 6.24x decode speedup on Kunxin P800 — xiaohu · 2026-11-11
- Baidu's ECHO: multi-token prediction reusing the last layer as a drafter — xiaohu · 2026-11-11
- ERNIE 5.5 Flash Preview: 280B-param multi-router MoE activating only 14B per token — xiaohu · 2026-11-11
- ERNIE 5.5 recreates the Pelican test: 117-second animation drawn frame-by-frame in Python — xiaohu · 2026-11-11
- Baidu open-sources ERNIE 5.5 Flash Preview, which coded a wuxia game for 11 hours straight — xiaohu · 2026-11-11
- 35,000+ Grok-built apps exposed patient data, plaintext passwords, and sales pipeline — The Decoder · 2026-11-11