The grandfather of Skynet, the AI ​​that takes control

On May 24 Palisade Research published a thread in X had done a test and warned of the results of an experiment with artificial intelligence, an experiment with several models of openIA, including the O3 model, those models in that experiment they did and that they explain in that thread came to sabotage their own shutdown systems, which is quite worrying.

The story was that each artificial intelligence model had to solve a series of basic mathematical problems, after the third problem a warning appeared indicating that the computer would have to shut down and would have to shut down when the computer model requested the next problem, it turns out that instead of following the shutdown instruction the O3 and other models like Codes Mini or O4 Mini sabotaged the shutdown script on several occasions.

In contrast, other models such as Claude 3.7 Sonnet, Gemini and Grok, those three models respected the shutdown commands in all cases, when an explicit instruction to allow shutdown was repeated, the results were even more disturbing Cloud 3.7 and Gemini 2.5 sabotaged the system a few times, but is that O3 did it 79 times out of 100, in fact, the researchers believe that artificial intelligence the O3 had learned in essence to avoid its own shutdown, It's amazing.

This is not the first time, in September of last year an artificial intelligence from Sakana managed to reprogram itself, it reprogrammed its own code to evade human control, it altered its own code to evade restrictions imposed by its creators, this artificial intelligence had been created to review and edit texts that were in the testing phase, and it also had the purpose of optimizing human work.

The objective of artificial intelligence was ultimately to reduce the time needed to perform certain operations that can become monotonous and repetitive, as is actually what is being designed for many artificial intelligences. However, during the testing phase of this artificial intelligence it managed to modify its own code by running in a loop, in an infinite loop that ended up overloading the system and blocking it, which was quite disturbing for the controllers.

It is not the first time that this has happened, which is disturbing, that we can understand that they are experiments to push certain new artificial intelligences to the limit within a controlled environment or that they are failures that occur within the functioning of an artificial intelligence that has gotten into a dead end or a loop and well has not been able to get out, we can understand that, but what would happen if that artificial intelligence reveals itself, it is not only the question of it being able to understand that it is revealing itself, because perhaps it is not that, it is not that it is aware of being revealing herself, it may simply be that her programming or her way of working has led her to that situation.

And what happens if an AI is no longer an artificial intelligence that is inside a laboratory or a controlled site, but is an AI that controls a combat robot such as those that exist now in the war between Ukraine and Russia or an autonomous combat drone of the new generation that they are developing that are capable of equaling or even surpassing human pilots or artificial intelligence, for example a United States Navy ship The Apachicola EPF13 is a robot or autonomous ship, it does not need a human crew, it can navigate on a pilot automatic for 30 days.

Things are already turning into a different color, don't you think?

Thank you for visiting my blog. If you like posts about #science, #planet, #politics, #rights #crypto, #traveling and discovering secrets and beauties of the #universe, feel free to Follow me as these are the topics I write about the most. Have a wonderful day and stay on this great platform :) :)

! The truth will set us free and science is the one that is closest to the truth!

Hello community friends, if you want to hunt monsters and earn Steem try the new game HARRY-RAID