Show HN: I made Jev to talk

Category: ai-ml

Tags: experimental, generative-text, decision-models, python, llm-alternative

Score: 6.8/10 (Innovation: 9, Technical: 7, Documentation: 8, Utility: 3)

Jev Talks is an experimental Python project that forces a decision-only API (Choice/Noul primitives) to generate natural language answers by iteratively scouting candidate words via multiple-choice and judging phrase continuations via yes/no questions. It's notable for turning a probability-output-only model into a word-by-word text generator, complete with a diffusion variant, self-grading rubric, and detailed cost/latency analysis.

Target audience: AI/ML researchers and hobbyists exploring probabilistic and non-autoregressive text generation

Repository: https://github.com/zie1ony/jev-talks · Python · 6 stars

View on Hacker News