Jev-Powered Character-by-Character LLM
Vogel says he made an LLM from first principles using Jev. For each character, it asks 29 yes/no questions to choose among a–z, space, comma, and period.
vogel@ryanvogel𝕏
i made an llm from first principles with Jev 29 yes/no questions per character: should the next key be a–z, space, comma, or period? highest probability gets append to it, then fed the updated text back in & repeat an autoregressive loop made out of a classifier https://t.co/OdiVS5MmM1
It appends the highest-probability option to the text, feeds the updated text back in, and repeats in an autoregressive loop built from a classifier.
Also filed under Tools & apps
- Search Telegram Channels by Topic
Sanyi built a tool using JEV that lets people search Telegram channels in his community by topic, question, or vibe. Matching channels rise up in seconds.
- Jev Omni MCP Build
eureka-pd/jev-omni-mcp is an MCP build for Jev.
- Jev Verbose Couscous Build
Beliato's `jev-verbose-couscous` is a build made with Jev.
- Build a Twilio IVR with Jev
This project uses Jev with Twilio to build an interactive voice response system.