Please call it Harness/LLM

Melanie Mitchell points out that nobody really uses LLMs by themselves anymore. What we call ChatGPT or Claude is an amalgamation of a model, post-training, safeguards, and an ever-more-elaborate software harness.

I’d just like to interject for a moment. What you’re referring to as an LLM, is in fact, Harness/LLM, or as I’ve recently taken to calling it, Harness plus LLM. An LLM is not a AI system unto itself, but rather another component in a fully functioning software harness made useful by the system prompt, context management, tool loops, safeguards and vital system components comprising a full AI system as defined by ISO/IEC 22989.

Through a peculiar turn of events, the whole system is often simply called an “LLM.”

Please use the technically correct term: Harness/LLM.

(The above is a dumb joke post based on the infamous “I’d just like to interject” copypasta attributed to Richard Stallman.)