Post #4725462 · Elektrine

2026-09-05 22:56 UTC

@gumnos@mastodon.bsd.cafe Yes, that's very clever. But destruction, no matter how uninformed the targeted victim, just doesn't sit well with me. I don't think stupidity should be punished; pity is a safer route. Doing harm is a slippery slope.

Replies (2)

@zzt@mas.to 2026-09-06 00:19

@VogueLaGalere@mastodon.online @gumnos@mastodon.bsd.cafe this you? https://mastodon.online/@VogueLaGalere/117167020575128079 doing harm is fine when it’s OpenAI doing it, then? fuck is wrong with you

Open ##4725463
@gumnos@mastodon.bsd.cafe 2026-09-06 00:52

@VogueLaGalere@mastodon.online Sometimes the only way to learn the dangers is to get burned. Despite countless people warning "fire is hot and will hurt you" (or "trusting LLMs with untrusted data/sources is dangerous"), playing with fire is the slippery slope. "pRoMpT eNgInEeRiNg" is a myth. Nothing about capital-E-Engineering applies to the process of crafting prompts. A prompt that works on one LLM might fail on another. Or might fail if the same LLM upgrades its model. Or even if it happens to run at a different time of day. It shares more resemblance with alchemy or astrology than Engineering.

Open ##4725465