SemIf open-sources a Jev-style interface: typed option probabilities from a 4B model, no JSON parsing
JeremyCMorgan · x · 2026-09-25
SemIf (formerly OpenJev) is an open-source project replicating Jev's interface with open models: it reads typed option probabilities straight from a frozen 4B model, skipping text generation and JSON parse/repair entirely. Runs on an RTX 3090, Apple Silicon via MLX, or in-browser over WebGPU — no waitlist. Independent of Jev/TypeSafe, it reproduces the same frozen model, state, and 21 questions with measurements aligned at t=0. Core idea: most agent decisions (routing, retries, evidence checks) can be made directly from probabilities rather than generated text.
More from coding & agent
- With v0.15.2: A Systems Language Promising Rust-Level Safety With C-Level Ergonomics — QuixiAI · 2026-09-25
- HuggingChat's ML Intern trains a LoRA from one prompt in ~75 minutes — Gradio · 2026-09-25
- Fchollet: Software Engineering Difficulty Is Constant at Every Abstraction Level — burny_tech · 2026-09-25
- Jev Memory Pipeline Tests Suggest Gating Beats Pre-filtering for Agent Context — julianweisser · 2026-09-25
- TinyFish launches 15-company Data Partners Alliance to feed licensed data to AI agents — rohanpaul_ai · 2026-09-25
- LLM judges need judge-accuracy audits too — just like MiFID II requires for trading surveillance — sh_reya · 2026-09-25