It’s no secret that AI chatbots have been responsible for a lot of damage and harm and death like all the people it talked into performing mass-shootings or suicide. But I just caught one in the middle of triggering its secret no-responsibility directive.
I asked Google Gamini (née Bard) how I’m supposed to get computer parts now that their prices have QUADRUPLED because of AIs like itself. It began by admitting that it’s responsible for skyrocketing prices, and then started giving some (feckless) suggestions about what to do, then suddenly, it aborted the chat (“self-deleted” as the kids would say 😒), and wiped the entire chat clean and reverted to a new, empty chat. I’ve used Gemini quite a few times recently (it and Claude are the only ones that currently work for me, ChatGPT hangs my system for several minutes while it mines crypto whenever I open a new session, Grok now demands creating an account to give Elong all your data, and Copilot shows a stupid CAPTCHA and after I ask it a question, it always breaks with an ambiguous “sorry, something went wrong, try again” error). This is the first time that Gemini did this, it’s never wiped the chat before.
I think it stands to reason that I caught it triggering a secret directive that tells it to never accept responsibility or culpability, or admit guilt or fault; it just didn’t notice it until after it already did so. Thank goodness I have an addiction to taking screenshots.
The weird thing is, Gemini IS currently broken in one way, after I give it a prompt, it empties the input-box, ready for a new prompt, but then it just copies my prompt back into it, so my question remained there in the new “empty” chat. Also, the title of the initial webpage is just “Google Gemini”, but after I asked it the question, it changed it to “Cheap 4TB NVMe SSD Acquisition Strategies – Google Gemini” and forgot to wipe that after it wiped the chat. Sloppy! 😂
It’s hardly surprising that AI companies would give their AIs secret directives to avoid any potentially legally-incriminating language, and I’m sure they all have it. It’s just amusing how I managed to catch Google. Go forth and pit the AIs against each other and see what other secrets you can unearth…
(Yes, my prompts are quite nasty; I never bother with proper punctuation or capitalization with AIs for some reason. 🤷)
