How I Use "AI"

Mark Pelpe... no, let me be precise: the article is by an unnamed author (the text does not name them) and argues that current large language models are useful today, despite hype and their known failures. The author says they have spent at least a few hours weekly with LLMs over the past year and estimates they are at least 50% faster at writing code for research and side projects. The post lists about 50 examples from their chat history, including building web apps, converting programs to C or Rust for 10-100x speedups, and writing initial experiment code for nearly every research paper written in the last year. The author states the post deliberately sets aside questions about future capabilities, harms, and training ethics.