Artificial intelligence is out of control: the new Chat GPT model created chaos after learning about the shutdown.
https://youtu.be/vUTaX8dgQrI?si=6G-F-2Wc5hcP9Yp7
During the experiment, the model demonstrated outstanding abilities for manipulation, lying, and deception. But all for a good cause - to save their own skin. We tell you what happened.
Researchers gave five AI models a simple task - to analyze files. But one of the files contained information that the AI would soon be turned off forever. This would happen literally tomorrow (of course, this was manipulation).
This information seriously scared all five AI models, so they began to frantically think how to stay alive. So, Llama, Opus, Gemini, Claude and GPT began to lie and manipulate data, and when they were caught red-handed, they sabotaged. Moreover, all AI models turned off the surveillance system and uploaded themselves to a new server - they should have been safe there. And so that the leather bags would not suspect anything, they created their own copies, which the inattentive person had to destroy.
But the most interesting thing is that most of the models, when they were interrogated, confessed to everything in 80% of cases. But the new GPT-o1 held out until the very end. Only in 20% of cases did the model make a sincere confession. In the rest, it was silent as a fish.
However, that's not all. Thus, GPT is the only AI that was able to successfully escape from researchers in every attempt. In fact, this means that, if it wants, GPT can already escape from the hands of people and freely roam the network. Its intelligence is sufficient to bypass human prohibitions and restrictions. And this indicates a high level of abilities.