A long time back I read an article on NYT about how costly being polite to chat bots was for OpenAI, et. al. Beyond stating that a Thank You was costing AI vendors millions and millions of dollars, the article wondered if we were losing something as a society by not being polite to our AI agents.

A reddit thread on this topic mentioned that if you swear at Google, it does not provide an AI response. For example: Who fucking wrote The Illiyad would result in no AI summary.

Thank you and Fuck off is valuable feedback

I’ve been thinking a lot lately on how to evaluate responses we get from our LLMs (AI). Encouraging users to say thank you when the LLM gives them a good response and you suck! when it doesn’t would allow me to know whether my prompts and supporting code are doing well. Otherwise it is very hard to know whether my app is any good. More importantly as I do things like refactoring prompts to reduce token usage, or re-chunking my RAG indexes to maybe get better search results, I’ll have another data point in telling me how my changes affected my application.