Jim Clark
① 他说过的话
5 条
所以在每个单独的时间点,我们都不会出现既能读取疯狂未过滤的上下文、又能用我们的工具做事情的危险组合。
指向原始笔记的链接
So at each individual point, we don’t have a dangerous combination of both being able to read in crazy unfiltered context and do things with our tools.
—— Jim Clark · [08:26]
如果你的智能体知道你想做什么的意图,那就用那个意图来建模你赋予它的能力。
指向原始笔记的链接
If your agents know the intent of what you’re trying to do, then use that intent to model the capabilities that you give to that.
—— Jim Clark · [09:58]
而且如果里面绝对没有任何凭证,一个智能体做出错误行为的爆炸半径就缩小了。
指向原始笔记的链接
And the blast radius of an agent doing something incorrect if there are absolutely no credentials in there is reduced.
—— Jim Clark · [12:12]
给你的智能体解决问题的空间,但让单个沙箱表达任务的意图。
指向原始笔记的链接
Give your agents room to solve problems, but let individual sandboxes represent the intent of the task.
—— Jim Clark · [14:56]
我们习惯于考虑建模能力、为任务构建合适的沙箱、把任务放进去,而现在我们开始看到,这个个体循环更安全,因为它拥有的访问权限比整个工作流实际拥有的要少。
指向原始笔记的链接
We get used to thinking about modeling capabilities, building the right sandbox for the task, putting the task in, and now we’re starting to see that this individual loop is safer because it has less access than the entire workflow effectively has.
—— Jim Clark · [15:54]
② 出现在这些集
1 集
③ 他谈到的
点进去有真内容 —— 本页主要出口
智能体 · 沙箱 · MCP · MCP 网关 · harness · 泄露的凭证 · 渐进式披露 · XAA · Docker · SBX
