^ Home

How I Feel About LLMs

Generative AI has been a thing for long enough that I've had enough time to formulate some opinions on its use:

• This website is a creative outlet for me and so I write 100% of its code (and of course content) myself.
• Coding assistants are fine to speed up development, as long as each change is kept small, a human remains in the loop (git add -p-style), and everything produced is reviewed at the time it's written. I dislike "agentic workflows" and do not vibe code. I think orchestration projects like Gas Town and ~fully LLM-generated projects like OpenClaw are crazy. I will always care about the code and its architecture, correctness, and testability.
• Using LLMs to generate prose, whether it's fiction, nonfiction, technical documentation, or even code comments, is an insult to the innate creativity of the human generating it, and the attention of any person expected to read it.
• All "art", "music", etc. produced by generative models is slop and debases both the community of people who produce art and music, and the person who generated it. In art, the process itself is one of the reasons for its existence, so even the most technically incompetent art is more valuable than anything machine-generated.
• I am deeply concerned with the downstream effects of model training and inference, and the political behavior of the US frontier labs. I believe that each person who chooses to use LLMs has the responsibility to run these models on their own hardware whenever possible.
• Machine learning in general is a very cool field and I totally support research and development in this broader area. My concerns are limited to the societal implications of how the current crop of generative tools are being used.

Last updated 26 Jun 2026