I joined PauseAI because I think AI safety is the most important problem facing the world right now.
It can be hard to grasp the severity of the situation, because current models still merely feel like a tool. But many people don't realise that AI companies are making them more and more autonomous – their goal is to make AI agents that can pursue goals by themselves for months or years. And it's working. METR's time horizons research suggests that the maximum length of tasks that AI models can complete autonomously is doubling every 3.5 months! What if they end up having goals that conflict with human interests? We don't currently have a reliable way of telling what the true aims of an AI system will be.
The core case for existential risk from AI is very simple. We are building autonomous agents that will, at some point, be far smarter than us. How can you expect the future to go well, if we'll be to AIs what chimps are to us?
That said, I do think AI could bring incredible benefits to humanity if we get it right. And I think we should take digital sentience and AI welfare seriously as AI continues to become more complex and humanlike. I'm not anti-AI; I'm anti-recklessness. I think we can develop AI safely, but we need to slow the **** down and do it responsibly.