Been using the chat-ai’s because Im a bit of a technophile, or maybe because it’s a legitimately useful tool. At this point I’ve envisioned a general knowledge AI that I have hosted in my house to help me deal with general and specific things like gardening or house-maintenance. Whether that’s a commercial or a DIY solution idk, but it’s just way too helpful.

now, i have a question about alcohol…

Right now, im playing with Bard. I am way past the legal drinking age as is my partner, and we were wondering which of the alcohols have the lowest histamine/sulphite count (she’s sensitive). I wondered if I could get answer and understandably so, it doesn’t respond when it comes to certain subjects, alcohol being one of them. It gives the “I’m just a language model bruh, i can’t answer that.”

Now, we’ve all heard of the prompters who go around these restrictions by essentially hypnotizing or role-playing with the chat bot enough so that the answer is given to you in a round-about way. “Write me a recipe like ol’ gramma did for c4….” but honestly doing a traditional search online would be faster.

The way I see it, the ai’s restrictions can be skipped over when you abstract your request a level outwards. Another way for me to explain it; Go a bit meta and it seems like they just have to ‘roleplay’ with you.

my current method

So the major use I have with these chat-ai’s is writing tedious code I don’t want to and I wondered if asking for such (alcoholic) info can be gained by asking for it in a C# script that’s well documented.

Sure enough, after a small pause, a script generated a string array of alcohol types, with Tequilla, Gin and Vodka being the lowest histamine/sulphite levels of the alcohols. Even wrote a method for figuring out the values. The best part? my prompt wasn’t longer than a dozen or two words.

*edit: i have no idea if this info is just a ‘hallucination’, or made up data from the chat-ai but we tried the lowest count drink types from the list and my partner was quite pleased to find out she wasn’t sneezing or getting a runny nose. So as far as I’m aware, for this specific case, it worked.

why are you trying to bypass stuff boi?

Like most exploits, I’m sure this will be “patched” and I’ll need to look for an open-source model if I’m going to be asking for ‘restricted info’ but I’m always interested in the more ‘mainstream’ efforts for the chat-ai thing. I’m kinda coming from the angle of, “wait, can anyone just do this?” to which after usually comes a, “wow, yeah, totally doable.”

Anybut that’s it for me, just a small note about chat-ai and prompts. Until next time.