TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Scott Aaronson Talks AI Safety (2022) [video]

2 点作者 quick_brown_fox大约 2 年前

1 comment

ggm大约 2 年前
I&#x27;m starting to think Aaronson is industrialising his commentaries to create his walk away wealth.<p>I&#x27;d be surprised if there are not deep roads in existing comp Sci, robotics and applied sciences to mine for safety principles. I don&#x27;t personally see problems in A.I. itself which haven&#x27;t been presaged by e.g. Norbert Weiner&#x27;s decision to withdraw from bomb science, or by the Oxford institutes work on ethics in computing and networks. Or dare I say it, the therac case, or the box girder bridge disasters freeman fox brought on.<p>&quot;Because a computer said it, it must be right&quot; is a terrible basis for decision making.<p>Quite recently for A.I. (within living memory, last time people talked up a new wave of research) might be the fuckups in expert systems such as medical school admissions software which encoded functional institutional racism and sexism: it optimised to mimic the bias inherent in &quot;manual&quot; admissions.<p>Healthy scepticism, air gaps and checking. Don&#x27;t believe anything Bender says right now. &quot;Good news people&quot; or not, Professor Farnsworth.