Das dunkle KI-Geheimnis, über das niemand spricht

But I don't have to lead an AI emotionally either. I can slap him left and right, say faster, further, higher and that's wrong, you complete idiot. And that works exactly the same way as if I said, please dear AI, today you don't seem to be feeling so well. Why do you think this

leadership isn't logical thinking? Because you could also say, "I need logical thinkers, those are physicists, mathematicians," but it's about leadership skills. At the beginning of 2025, a very controversial study was published about the collaboration between the University of Zurich and the psychiatry department in Zurich. And what they discovered is that AI is

traumatizable. This means that if I exchange information with AI about war events, for example, then [snorting] it has been shown that within this context window in which one acted with the AI, uh, the performance, when one then set any tasks, systematically decreased. And what this study has also shown is that with the therapeutic methods

we apply to people, we can get them back to their full capacity. Now, one might say that this is somehow esoteric and totally crazy. Not really, Ben, because we trained the AI on billions of data points of human behavior, a large part of the internet and a few other sources. Um, and what the AI ​​sees

is that when people talk about war or traumatic topics in forums or similar places, they react in a certain way. This has also been seen in other effects . There was, I think it was in 2024, this AI winter where suddenly everyone realized it was totally weird, I was giving prompts and only getting short

results, had the AI become lazy and so on, and the AI ​​researchers could n't explain it at all, they said, is it because people produce less output in winter, because we're not as productive, or is it a seasonal fluctuation in the data centers right now? Um, that could n't really be resolved, but that's what