I explored recent AI updates including Microsoft's Everything of Thought, fine-tuning datasets without inputs, and Tamil-Llama. I also checked out Voyager's Minecraft agent, Langchain's evaluators, and the power of using Pydantic to unify code, data, and text.
I learned about GPT Vision for calendar extraction, the limitations of RAG versus fine-tuning, and using LlamaIndex for hierarchical retrieval. I also found tools for JSON repair, LLM evaluation metrics, and running local models with Ollama.
I discovered Julius.ai for data queries and AutoGen for multi-agent systems. I also explored high-performance LLM inference using vLLM and DeepSpeed, alongside architectural patterns and critiques of the Foundation Model Transparency Index.
I explored various LLM tools, including llamafile for single-executable local models and marker for PDF-to-markdown conversion. I also looked into RAG tuning strategies, PII detection libraries like Presidio, and intuitions regarding scaling and emergent model behaviors.
A familiar PyCon topic became interesting again once the author reframed it as a live experiment in getting AI to generate code for Minecraft.
Reverse-engineering Minecraft's websocket interface turned a private obsession into a public guide for building bots and tools in Bedrock without official documentation.