研究人员提出了一种新的AI对齐方法，称为“价值泛化”，旨在创建能够可靠地将人类价值观和偏好扩展到新情况的AI。这种能力被认为是当前AI系统的一个关键缺失环节，因为AI在遇到训练数据之外的场景时常常会失败。提出的方法涉及将AI的道德概念与经验概念绑定，使其道德能够随着其能力一起成长和适应，从而可能实现本质上更值得信赖和可控的“预对齐”AI。 AI
影响 这项研究可能带来更可靠、更值得信赖的AI系统，使其能够在新颖的情况下安全运行，从而解决了AI开发中的一个关键挑战。
排序理由 该集群讨论了一种新颖的AI对齐研究概念，该概念在一系列学术/研究论坛的帖子中提出。
AI 生成摘要 · Google Gemini · 来自 6 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
<p><span>When we get explicit strong generalisation to work (see </span><a href="https://www.lesswrong.com/posts/58zFSWp8Tmxij6ckK/value-generalisation-i-an-r-and-d-program"><span>the first post</span></a><span> on the matter and </span><a href="https://www.lesswrong.com/posts/TZ…
<h1><span>A human superpower hidden from even ourselves</span></h1><p><span>I though GPT 3.5 was on the verge of Artificial General Intelligence (AGI). It certainly seemed that way – it could combine and extend ideas in ways that were far beyond narrow rigid computing. Sure, it h…
<p><span>I’m looking for people, advice, critiques, and funding to build a research program on value generalisation – the ability of an AI to correctly extend human values and preferences to situations neither it nor we have seen before. My ongoing research has become convinced t…
<p><span>When we get explicit strong generalisation to work (see </span><a href="https://www.lesswrong.com/posts/58zFSWp8Tmxij6ckK/value-generalisation-i-an-r-and-d-program"><span>the first post</span></a><span> on the matter and </span><a href="https://www.lesswrong.com/posts/TZ…
<h1><span>A human superpower hidden from even ourselves</span></h1><p><span>I though GPT 3.5 was on the verge of Artificial General Intelligence (AGI). It certainly seemed that way – it could combine and extend ideas in ways that were far beyond narrow rigid computing. Sure, it h…
<p><span>I’m looking for people, advice, critiques, and funding to build a research program on value generalisation – the ability of an AI to correctly extend human values and preferences to situations neither it nor we have seen before. My ongoing research has become convinced t…