MICA: an open-source LM with no neural network, just learned cellular automata rules
Silver_Employ2617 · reddit · 2026-10-01
MICA is a small language model with no neural network at inference — a grid of integer cells updated by learned rewrite rules. The project now ships two browser-playable open-source models: word-level Flame-W (34 MB, 20.7% next-word top-1) and byte-level Ember (4.5 MB, 1.86 bits/byte), with 139 conformance tests, research notes, and honest negative results. The key finding: MICA's effective context is only 4-8 bytes versus 32 for a small transformer on the same data, explaining why it loses the thread. The author built it alongside Claude, Codex, and Grok coding agents with strict reproducibility rules (model hashes and eval sets per result). Personal and research use only.
More from coding & agent
- Google Cloud's Agent Clinic: build an automated eval suite for a LangGraph agent in 60 min — LangChain · 2026-10-01
- GitHub's HydraFusion hits VS Code and Copilot app: multi-model routing at lower credit cost — DanWahlin · 2026-10-01
- His always-on agent cut his rent by $1000 and fights bills without him knowing — tech__unicorn · 2026-10-01
- Box CEO Aaron Levie: the AI deployment layer is a massive untapped opportunity — annbordetsky · 2026-10-01
- LangGraph adds typed response schemas for interrupts, making human-in-the-loop type-safe — LangChain · 2026-10-01
- Vibe-coded with Astra and Opus: web Rummy 500 game rebuilt in Godot for desktop — AIandDesign · 2026-10-01