AI

Nothing on this site or anything written in my voice on the internet was generated by AI. I treasure writing and the internet for thinking, self-expression, and connection with other humans. I am committed to using my own voice when writing anywhere on the internet and don't foresee that changing in the future.

Here's how I do use AI:

• I'm a power user of the Cursor code editor. I recently switched Cursor's underlying model from OpenAI's gpt-4o-mini to claude-3.5-sonnet at the suggestion of a few coders I respect, to great effect so far. I also started tinkering with a .cursorrules config file (docs) which has improved output. I've found that Cursor: (1) helps me focus more deeply on valuable work and less on bikesheddy technical wormholes, (2) reduces my hesitance to dive into new languages, packages, frameworks, etc., (3) dramatically mitigates the grindy parts of coding, and (4) increases my happiness. For reasons that feel obvious to me, I heavily revise and repurpose AI-generated code to make it more readable, maintainable, and generally fit into whatever I'm working on. In short, while AI helps me write code, I take responsibility for molding it into something that's useful in the real world and high quality for my fellow nerds.
• I use Raycast Pro, which has some AI chat tools that I find helpful in limited ways. The OpenAI and Anthropic extensions are useful for short answers to simple questions. I occassionally use Raycast's AI chat to help me think divergently about things I'm writing or working on. I never use the output directly in anything that I write. Overall, I find the actual responses from AI to be much less valuable than the process of thoughtfully identifying what I'm trying to make sense of and formulating clear questions to the AI. By the time I've prepared my submission to the pattern-matching virtual dumbass, the solution tends to have presented itself. Which, I suppose, is the value of a rubber duck.
• I've tried several AI-integrated search engines (like Perplexity) but am sort of fizzling out on them. Honestly, the value hasn't exceeded the drawbacks for me. When I'm searching on the internet, I'm looking for a diverse variety of human-curated sources, not a single reductionist robo-answer. I'm looking for things from people I trust, not the mathematical mean of what's on the internet. My searches are more about starting points, not final answers, and Perplexity hasn't really improved much on what I can do with Google. Searching, for anything really, implicates a social context that I'm not confident AI will ever be able to replace for me.

Finally, one contrarian AI take I have is that "hallucinations" are immensely helpful. They heighten my diligence, force me to be truly engaged with whatever I'm working on, and, ultimately, reveal tons of insights. I think the key is acknowledging that I'm working with a pattern matching algorithm and not actual "intelligence."