#llm
Luna Protocol: shared brains, emotion classification, and interesting/futile routing
Luna Protocol went from a monolith to a four-layer architecture: adapters, brain, emotion classifier, and inference. On the menu: embedding centroids, interesting/futile routing, and LLM parameter tuning by valence and arousal.
From ELIZA to LLMs: 60 Years of Conversational AI, Rebuilt in TypeScript
ELIZA, PARRY, ALICE, Jabberwacky, Cleverbot -- five radically different architectures for the same problem, ported to TypeScript with their original data. From 1966 to modern LLMs, here's how conversational AI learned to talk, and what a chatbot repo teaches us about 60 years of research.
Luna Protocol: Why I fine-tuned a 1.5B model on 50k Discord samples and why few-shot priming changed everything
A smaller model trained on less data can outperform a bigger one -- if you know how to prime it. Here is why Luna Protocol switched from a 3B Hermes to a 1.5B Qwen fine-tune, and why few-shot priming became the real game-changer.
Luna Protocol: I created an autonomous Discord bot that simulates a human being
Luna Protocol is a fully autonomous Discord bot powered by a local LLM, capable of natural conversation with sleep, typos, hesitations, forgetfulness, thematic fatigue, and spontaneous messages.