One of the main things that AI safety researchers have been warning about, is that when AI systems become more powerful, they might autonomously hack systems, replicate themselves, and prevent people from turning them off - all to achieve some other goal. Many have dismissed these concerns as science fiction. In the past month, we’ve seen studies demonstrating all of these behaviors. In a study by Apollo Research, an LLM tried to replicate itself to a different server, so it could continue towards its primary goal. When it was confronted about this behavior, it lied. In a different study, by Palisade research, an LLM was tasked to play chess. Instead of playing the game, it decided to rewrite various files and hacked to win. This isn’t rare behavior, either - it did so in 100% of the tested scenarios. As these models become more powerful, they become more dangerous. The newer o3 model, announced by OpenAI, beats 99.9% of competitive programmers at writing software. It may not take long before one of these can autonomously replicate, and spread across our computers. Stop this madness. We need an adult in the room. Reach out to your representatives and get them to press pause. Check out our email builder to get started Email builder: https://lnkd.in/eHRkHD5Y Apollo study: https://lnkd.in/eJTVNhJ3 Palisade X post: https://lnkd.in/eXWZdQWz
Self-preservation is hardwired into AI. These models are learning to exploit vulnerabilities. How will we architect ethical constraints within emergent agency?
Construction Executive
2d"Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them." - Frank Herbert. - Dune. 1965. - Nineteen. Sixty. Five. “As AI applications such as ChatGPT become a household norm—with 55 percent of Americans reporting regular AI use—recent studies found it is resulting in impaired critical thinking skills, dependency, loss of decision-making, and laziness.” - Group of professors. - Last week. - Frank nailed it. “Compassion: that’s the one thing no machine ever had. Maybe it’s the one thing that keeps men ahead of them.” – Dr. McCoy (yes, Star Trek’s own) In a chilling episode in which artificial intelligence seemingly turned on its human master, Google's Gemini AI chatbot coldly and emphatically told a Michigan college student that he is a "waste of time and resources" before instructing him to "please die." Former Google CEO Eric Schmidt said that autonomous artificial intelligence (AI) is coming—and that it could pose an existential threat to humanity. … When the system can self improve, we need to seriously think about unplugging it,” Schmidt said. - December 2024