# Deep Dive Relevance Keywords # Define keywords and their weights for scoring entries # Weight tiers: High (3.0x), Medium (1.5x), Low (0.5x) weights: high: 3.0 medium: 1.5 low: 0.5 # High-priority keywords (critical to Hermes/Timmy work) high: # Framework specific - hermes - timmy - timmy foundation - langchain - langgraph - crewai - autogen - autogpt - babyagi # Agent concepts - llm agent - llm agents - agent framework - agent frameworks - multi-agent - multi agent - agent orchestration - agentic - agentic workflow - agent system # Tool use - tool use - tool calling - function calling - mcp - model context protocol - toolformer - gorilla # Reasoning - chain-of-thought - chain of thought - reasoning - planning - reflection - self-reflection # RL and training - reinforcement learning - RLHF - DPO - GRPO - PPO - preference optimization - alignment # Fine tuning - fine-tuning - finetuning - instruction tuning - supervised fine-tuning - sft - peft - lora # Safety - ai safety - constitutional ai - red teaming - adversarial # Medium-priority keywords (relevant to AI work) medium: # Core concepts - llm - large language model - foundation model - transformer - attention mechanism - prompting - prompt engineering - few-shot - zero-shot - in-context learning # Architecture - mixture of experts - MoE - retrieval augmented generation - RAG - vector database - embeddings - semantic search # Inference - inference optimization - quantization - model distillation - knowledge distillation - KV cache - speculative decoding - vLLM # Open research - open source - open weight - llama - mistral - qwen - deepseek # Companies - openai - anthropic - claude - gpt - gemini - deepmind - google ai # Low-priority keywords (general AI) low: - artificial intelligence - machine learning - deep learning - neural network - natural language processing - NLP - computer vision # Source-specific bonuses (points added based on source) source_bonuses: arxiv_ai: 0.5 arxiv_cl: 0.5 arxiv_lg: 0.5 openai_blog: 0.3 anthropic_news: 0.4 deepmind_news: 0.3 # Filter settings filter: min_relevance_score: 2.0 max_entries_per_briefing: 15 embedding_model: "all-MiniLM-L6-v2" use_embeddings: true