From the release of ChatGPT in 2022 until this year, I used LLMs almost exclusively as augmented search engines. They replaced much of my Stack Overflow and Google use, but did not otherwise make a large impact on my programming work or my life.
Since May, the range of tasks that I entrust to LLMs has expanded significantly. This is partly a function of advances in AI technology (more capable models, reasoning ability, tool use) and partly a function of me paying more attention.
I spent much of July testing LLM capabilities, exploring different AI-based products, and building my own tools for working with LLMs. This series of essays and notes is the result.
- A mental model of LLMs (300 words)
- My LLM tooling (300 words)
- My personal LLM evals (1,300 words)
- Can an LLM write literary prose? (800 words)
- LLMs outside of programming (300 words)
- CSS recipes from Claude Code (300 words) ∎