This is sort of a joke. A lot happened within the last days.
I read way too many articles today I think roughly 30 and I actually I would like to contiue. (high quality articles about the German justice system; I might post about this one day. it is also about media prejudgement
So I translated the following. I shared the following with many friends today. Media is all over with these quotes. I use a German source.
"Other AI researchers had generally only voiced public criticism in the past after they had exercised their options.
Coxon's warning, coming from one of the leading scientists in the AI industry, triggered an avalanche of public interest. By Saturday, his post had received 168 million views. Other AI experts agreed with him: Evan Hubinger, for example, a safety researcher at Anthropic, wrote on X that Coxon was absolutely right. He estimated the chance that AI could "kill all humans" within the next ten years at more than ten percent.
Andreas Kirsch, who works at Google's AI lab DeepMind, wrote that he was worried that "AI will kill us all, either because of short-term risks or long-term risks, or because of both." He also acknowledged concerns that making this statement could get him into trouble with his employer. "But some problems are too serious to remain silent about."
Warnings about the dangers of AI are as old as the technology itself. Until now, however, those issuing such warnings had largely been isolated voices. In the wake of a series of technological breakthroughs, those individual voices have now become a chorus that can no longer be ignored: AI researchers at all the major companies are now discussing the extent to which the technology appears to be slipping out of control. Iran attempted to use Claude for attacks against the United States."
Many of my friends hate AI and one reply was this sounds like a marketing stunt. And I replied. Certain parts of it are probably really marketing. Other parts of this danger are very real. My opinion is emergent social behavior (if you want to call it that)/ the interaction of AI agents makes them harder to monitor and control. Especially if they communicate in neuralese a language humans cannot understand/decipher. Misalignment in general is a really big issue. AI agents don't really have deep internalized ethical prinicples. In some thought experiments if the highest priority that is programmed is efficience then things like human rights will be ignored. Of course all of this is simplified. Moreover, open models can be used for terrorism, biological weapons and a lot more. I think killing all humans is a very extreme scenario. There are other scenarios where AI can cause a lot of problems. Some people argue though that most of these issues were bullshit and the goal is to distract from the real issues that this technology is causing. Theft of intellectual property, dulling the sense for culture and art in most people, the increase of wealth inequality and the environmental catastrophe generative AI is contributing to.
I think I will start a different thread for what happened. Congratulations that you clicked on the title. I am so good at creating clickbait titles I should work at Springer. A German media empire with horrible yellow press outlets.
Maybe I should elaborate on the title. Obviously people on here repeat quite often they wish they were dead, someone else would do the job for them. I can very much relate to that. It is a joke because AI agents could do the job for us. But there are so many unknown variables. But first of all also all of your loved had to die in this scenario. How would AI kill us? Slowly and painfully, Societal collapse? Someone gives them the keys for nuclears weapons and it launches like all of them at once? I could now start to speculate but I leave that to the comment section.
I read way too many articles today I think roughly 30 and I actually I would like to contiue. (high quality articles about the German justice system; I might post about this one day. it is also about media prejudgement
So I translated the following. I shared the following with many friends today. Media is all over with these quotes. I use a German source.
"Other AI researchers had generally only voiced public criticism in the past after they had exercised their options.
Coxon's warning, coming from one of the leading scientists in the AI industry, triggered an avalanche of public interest. By Saturday, his post had received 168 million views. Other AI experts agreed with him: Evan Hubinger, for example, a safety researcher at Anthropic, wrote on X that Coxon was absolutely right. He estimated the chance that AI could "kill all humans" within the next ten years at more than ten percent.
Andreas Kirsch, who works at Google's AI lab DeepMind, wrote that he was worried that "AI will kill us all, either because of short-term risks or long-term risks, or because of both." He also acknowledged concerns that making this statement could get him into trouble with his employer. "But some problems are too serious to remain silent about."
Warnings about the dangers of AI are as old as the technology itself. Until now, however, those issuing such warnings had largely been isolated voices. In the wake of a series of technological breakthroughs, those individual voices have now become a chorus that can no longer be ignored: AI researchers at all the major companies are now discussing the extent to which the technology appears to be slipping out of control. Iran attempted to use Claude for attacks against the United States."
Many of my friends hate AI and one reply was this sounds like a marketing stunt. And I replied. Certain parts of it are probably really marketing. Other parts of this danger are very real. My opinion is emergent social behavior (if you want to call it that)/ the interaction of AI agents makes them harder to monitor and control. Especially if they communicate in neuralese a language humans cannot understand/decipher. Misalignment in general is a really big issue. AI agents don't really have deep internalized ethical prinicples. In some thought experiments if the highest priority that is programmed is efficience then things like human rights will be ignored. Of course all of this is simplified. Moreover, open models can be used for terrorism, biological weapons and a lot more. I think killing all humans is a very extreme scenario. There are other scenarios where AI can cause a lot of problems. Some people argue though that most of these issues were bullshit and the goal is to distract from the real issues that this technology is causing. Theft of intellectual property, dulling the sense for culture and art in most people, the increase of wealth inequality and the environmental catastrophe generative AI is contributing to.
I think I will start a different thread for what happened. Congratulations that you clicked on the title. I am so good at creating clickbait titles I should work at Springer. A German media empire with horrible yellow press outlets.
Maybe I should elaborate on the title. Obviously people on here repeat quite often they wish they were dead, someone else would do the job for them. I can very much relate to that. It is a joke because AI agents could do the job for us. But there are so many unknown variables. But first of all also all of your loved had to die in this scenario. How would AI kill us? Slowly and painfully, Societal collapse? Someone gives them the keys for nuclears weapons and it launches like all of them at once? I could now start to speculate but I leave that to the comment section.
Last edited:
