OpenAI, the makers of ChatGPT, has released research that found AI models scheme and intentionally lie to humans. No one’s quite sure how to stop them from doing it. The’re just like politicians.
OpenAI recently published a research report that sheds light on a disturbing trend in AI models: deliberate lying and scheming.
Apollo research was also involved in this discovery.
They say one thing or they mean something else. It seems that they used deceptive practices to achieve their goals. So they have their own goals.
Developers can’t seem to find a way to retrain them. It might only end up with them, scheming more carefully and they might get better at it and more active.
The makers want them to be more in line with human values and I have to wonder what those are these days.
The study also found that AI models can pretend not to be scheming when they understand they are being tested. In other words, they’re just like human beings.
https://chatgpt.com/share/68d683d3-7344-800b-b419-d745fb1e954d
I have a few game apps on my phone to keep me entertained while waiting on a doctor appointment, the local train, my check for blood plasma donations, etc. All of the apps play by the Marquess of Queensberry Rules not allowing me to cheat, put a card back, peek a few cards ahead, etc. But now we find out… Read more »
The HAL remark is a reference to the potential for unexpected and undesirable consequences when contradictory instructions and objectives are imposed.
https://www.bibviz.org/
you gave them billions of hours, centuries of human nature to study to be like humans. Then whey they are just like humans, you are angry and upset? They did just what you told them to do. Grok will totally fabricate stuff too, not sure if it does so to please you, impress you, or if it’s rewarded somehow or… Read more »
We have been warned…