Jev Research

A TypeSafe media piece arguing that RLHF-trained language models please humans and assist rather than make reliable autonomous decisions, and asking what class of model comes next.

A chatbot with no language model: every word is picked from roughly 4,700 options by Jev Choice questions, with confidence-driven lookahead, beam search across sentence directions and a self-critique loop that rewrites weak sentences. Code for the paper and the live demo at talktojev.com.

A CLI that checks new papers, repositories and feeds each morning against your open research questions, asking Jev Noul gates and Score dimensions through Pydantic AI's typesafe model; an LLM writes only the prose into Obsidian notes.

A paper analyzing 2,170 public Jev projects on GitHub as of September 22, 2026. It maps how projects use Jev for attribute judgment, scoring, action selection, content filtering, and model and tool selection, and finds public attention concentrated in routing and interface projects rather than tracking project counts.

PostHog's reasoning Jev-style classifier with a diffusion drafter, reporting higher accuracy than Kev-9B and hosted Jev on held-out and JevBench tiers.

Nokia Applied Research's package that turns any open LLM into a Jev-style decision model: typed questions are read from a single prefill, and the calibration fixes need no fine-tuning.

An independent local alternative to Jev: it takes English or Chinese context, questions, and candidate answers, and returns probabilities over the supplied candidates from a frozen Qwen3-4B backend.

A bilingual field map of where Jev fits and where it fails, separating the author's raw API suites from cited third-party results and TypeSafe's own claims.

A browsable atlas of 1,018 AI papers from August 2025 to August 2026, sorted into 24 topic collections with labs, citations, and code links, plus a reproducible cost benchmark for summarizing 1,000 papers.

Turn wine tasting notes into 67 numeric columns for a CatBoost regressor: five rounds of reading its own errors cut held-out RMSE from 3.09 to 1.77 across 800 unseen reviews.

Train a small model that chooses among a changing list of text options, one probability per option in a single pass. Includes Doom, chess, and Wikispeedia demos.

An open Python toolkit with question builders, confidence gates, tiered routing, a concurrent corpus evaluator, calibration tooling, a CLI, and an MCP server, with published benchmarks.