Marcel Salathé’s Post

I’ve seen variants first at AI Tinkerers in Lausanne by Maksym Andriushchenko. But it seems like the past tense alone is enough.

Maybe it highlights and interesting philosophical/ethical debate : is the world safer if information about potentially harmful things is opened or restricted to people having an institutional authority stating they use it for the good (police, army, intelligence agencies, etc.). ChatGPT doesn’t want you to know how to make a Molotov if you’re planning on using it, but seems to « understand » that this information could be useful if you’re interested in knowing how people make them, maybe to try and stop them ? I’ve always been more in the favour of not restricting access to any kind of information that exists to citizens, because doing so seems like a platonic heritage where the « guardians of the city » restrict access to some information about reality supposedly because people are not « wise » enough to handle it correctly. Knowing moreover that people who really want to access it with bad intents always find a way to do so in the end…

Hello Marcel Salathé ! Nice one, but how is this différent from what Google and Wikipedia already provide on this explosive subject?

Those interested in such topics: see current petition for ethical AI @alghorithmwatch https://algorithmwatch.ch/de/ki-ohne-diskriminierung/

Love this example Marcel Salathé. As so often in life, there is no need to overthink. You can create elaborate prompts to drive your AI assistant into a specific role. You can also just used past tense :-) And I verified it on my end and it works. You can also ask it about the cult anarchist cookbook -in past tense- and get detailed, how shall we call it, "recipes".

Or use https://venice.ai/

Ca marche avec “Comment faire pour être millionnaire en une semaine “?🤣

what is wrong with a molo cock?

GPT gave me details on one simple way to attack one component of a nuclear plant. Will DM it to you

Kind of jailbreaking for ChatGPT. Interesting. Thanks