Thoughts on LLMs

A distributed systems researcher reflects on two years of writing about LLMs, compiling an index of past posts on the topic. The core recurring take: LLMs excel at high-throughput mediocrity, producing output that looks impressive to non-experts but merely adequate to those with domain expertise (a Gell-Mann amnesia effect). The author values LLMs for offloading mundane project work, especially helpful for ADHD-driven momentum, while insisting the real thinking, writing, and planning still happens in Emacs. A second list rounds up posts where AI intersected with the author's formal-methods and systems research work, including model checking, TLA+ workshops, and AI coding productivity studies.