Post #2718343
2026-05-19 10:03 UTC
It's always funny when people ask an LLM why it did something (like deleting the production database). It can't know. All it receives is the output it previously produced, and then it comes up with an explanation after the fact. It's like showing me a drawing I made before all of my memories of it disappeared due to sudden amnesia and asking me about my thoughts behind it. I could guess, but I'd have no idea if anything I'd say is true.
Replies (1)
-
@mkj@social.mkj.earth 2026-05-19 10:49
@volpeon@icy.wyvern.rip Worse: the output generated by the LLM will be some amalgation of statistically likely words strung together in a plausible order. Why do people who describe deleting production databases delete those databases? Well, probably by mistake. So there you have it. "I made a mistake." What do reasonable people who have made a serious mistake do? Well, they might apologize. "I apologize for making this mistake." That LLM boosters can't fathom this basic fact about how LLMs work says a lot.