Microsoft’s AI-powered “Bing Chat” service doesn’t seem to be reacting well to attempts to hack it <https://arstechnica.com/information-technology/2023/02/ai-powered-bing-chat-loses-its-mind-when-fed-ars-technica-article/>. It has started arguing with users, and even showed signs of being upset when they mention its code name “Sydney”.
One user asked it about prompt-injection attacks, to which it replied that it was immune to them. Then, when pointed to the previous Ars Technica article detailing exactly such an exploit, it basically went into full denial mode, claiming that “the article is published by a source that has a history of spreading misinformation and sensationalism”. Screenshots of other less-than-wholesome interactions are derided as “not authentic” and “edited or fabricated”.
I think the psychological term for such flat-out denials of reality is “psychosis”.
And looks like we finally have our real-life HAL 9000, just about 31 years late. Just make sure it is never connected to any life-support systems...
On Monday night, Ian gave a presentation on ChatGPT: https://www.youtube.com/watch?v=qEqjfqja72M Though it was aimed at writing (Python) code, we also had a general play with it towards the end. And, of course, tried some injections as well. Cheers, Peter -- Peter Reutemann Dept. of Computer Science University of Waikato, Hamilton, NZ Mobile +64 22 190 2375 https://www.cs.waikato.ac.nz/~fracpete/ http://www.data-mining.co.nz/