A diverse AV stack that combines a modular stack and NVIDIA Alpamayo reasoning VLA models for algorithmic AI safety.
大多数人认为自动驾驶安全应该依赖于单一、确定性的算法来确保可靠性，但作者认为结合模块化堆栈和推理VLA模型的多样化方法才能实现真正的算法安全。这种观点挑战了行业对单一'最佳算法'的追求，提出多样性本身就是安全策略的一部分。
2 Matching Annotations
- Last 7 days
-
www.nvidia.com www.nvidia.com
-
- Dec 2022
-
pluralistic.net pluralistic.net
-
Alas, lawmakers are way behind the curve on this, demanding new "online safety" rules that require firms to break E2E and block third-party de-enshittification tools: https://www.openrightsgroup.org/blog/online-safety-made-dangerous/ The online free speech debate is stupid because it has all the wrong focuses: Focusing on improving algorithms, not whether you can even get a feed of things you asked to see; Focusing on whether unsolicited messages are delivered, not whether solicited messages reach their readers; Focusing on algorithmic transparency, not whether you can opt out of the behavioral tracking that produces training data for algorithms; Focusing on whether platforms are policing their users well enough, not whether we can leave a platform without losing our important social, professional and personal ties; Focusing on whether the limits on our speech violate the First Amendment, rather than whether they are unfair: https://doctorow.medium.com/yes-its-censorship-2026c9edc0fd
This list is particularly good.
Proper regulation of end to end services would encourage the creation of filtering and other tools which would tend to benefit users rather than benefit the rent seeking of the corporations which own the pipes.
-