I explored how httpretty mocks Python HTTP libraries and learned about Japanese ancestral worship practices. I also researched GPT-4o's use of CNNs and OCR for image embeddings and visualized sinusoidal series as geometric spirograms.
The LLM cost-quality frontier keeps shifting quickly, and OpenAI's August 2024 price cuts briefly restored GPT-4o and GPT-4o mini to the best-value position.
I tested LLMs using Caesar cipher prompts, compiled a list of cheap cloud GPU services like Runpod, and learned how JSR handles package documentation. I also found that averaging embeddings is useful for processing long document inputs.
I evaluated AI detectors like GPTZero, finding them more accurate than Undetectable.ai. I also learned about CoVA for OCR-based scraping and the necessity of maintaining short, reliable feedback loops when using LLMs to write or maintain code.
I explored using Pyodide for browser DOM access and PyMuPDF4LLM for converting PDFs to Markdown. I also compared AVIF to GIF compression, experimented with Opus audio encoding, and researched Anthropic’s contextual retrieval methods for improved RAG performance.
I tested AI coding tools like Aider, explored MCP servers for automating SaaS tasks, and compared the latest models like Gemini 2.5 and GPT-4.5. I also delved into CRDTs, spatial hash indexing, and why I now prioritize evals for LLM projects.