A running list of what I'm actively exploring in AI. Completed items are marked with strikethrough.

• MIT OpenCourseware 15.773 Hands-On Deep Learning
• Strategies for Ed-Fi API tool use by LLMs (MCP or other)
• Current strategies for hallucination detection
• Hallucination detection: LLM-as-judge
• Learning to critical read and analyze LLM Benchmarks & Leaderboards
• LLM output validation / evaluation
• Hallucination detection: Semantic entropy
• Hallucination detection: Beyesian estimation of semantic entropy
• LLM tool use mechanics
• Token efficiency with tool calling
• LLM-hosted tool calling costs and implications
• Forcing JSON/other schema responses by forcing LLM to use 'fake' tool
• Security considerations with LLM tool use
• MCP Server primitives: tools, resources, prompts
• Nuances between function/tool calling and various packaged LLM features (agents, memory, etc.)
• Tokenization, Byte Pair Encoding
• Prompt Caching
• Token pricing: input, output, cached, etc.
• Token count/cost difference by format (JSON, YAML, code, prose, etc)
• Designing LLM conversations for efficient cost
• Developing and testing intuition on model selection (Claude haiku vs sonnet vs opus)
• LLM batch processing
• Anthropic's Contextual Retrieval offering
• RAG pipeline, RAG types (traditional with embeddings, Prompt RAG / table of contents pattern, ELITE, full text search, Graph RAG, etc), vector databases, semantic search
• pipeline: query -> decompose -> run each -> rerank summed results
• Generating embeddings via Azure AI tools
• Embeddings models, tradeoffs, and how to choose
• Semantic math with embeddings