Cellular-automaton LM MICA v0.3 adds 64-word memory, boosts context use 30x
Silver_Employ2617 · reddit · 2026-10-04
An experimental language model built on a cellular automaton — no Transformer, no floating-point network — got a 674 KB integer memory module looking back 64 words. Context-use at 8-word distance jumped from 0.0006 to 0.04, validation loss improved 5.309→5.268 bits/word, and Tiny Theory-of-Mind rose 28.15%→30.95% (beating 6 of 36 listed models). Key negative result: blind 50-prompt tests show sentence quality didn't improve — memory reaches the output but not the automaton's internal dynamics. Code, models and failed experiments are public in Vovala14/Mica-Ai; implementation is largely done by Claude/Codex coding agents with strict result tracking.
More from coding & agent
- Dialkit macOS ships open source: tune iOS interfaces live from your Mac — mikelikesdesign · 2026-10-04
- KMP: recovering project decisions and evidence across Claude and Codex via MCP — Mountain-Raise-4556 · 2026-10-04
- (Lean)DOOM: DOOM fully rewritten in the Lean proof assistant, with formal proofs included — akbirthko · 2026-10-04
- Jin: a minimalist coding agent that swaps MCP/plugins for prompts and bash — aldapsiger · 2026-10-04
- Kepler hits server-verified 100 on all 25 ARC-AGI-3 games with Opus 5, for $777.72 — rohanpaul_ai · 2026-10-04
- Agent hit 100 on ARC-AGI-3 game by reading its source code; clean rerun scored 46.91 — rohanpaul_ai · 2026-10-04