① 他说过的话

15 条

交付给患者,会把你通常发布时依赖的那些常规安全网全部拿走。
So shipping to patients takes away the normal safety nets you would normally ship with.
—— Jared Joselowitz · [00:58]

指向原始笔记的链接

而且非常重要的一点是,模型卡救不了你。
And very importantly, the model card won’t save you.
—— Jared Joselowitz · [01:17]

指向原始笔记的链接

你不能像一些模型供应商那样声称某个基准达到了 92%,在事后审查中这不是一种辩护。
You can’t claim like some model vendors said that they have 92% on some benchmark. It’s not a defense at a post-instant review.
—— Jared Joselowitz · [01:21]

指向原始笔记的链接

你不能撤销一通电话。Dora 一旦说了,话就说出去了,没有回滚。
You can’t undo a call. Once Dora says it, it’s been said, and there is no rollback.
—— Jared Joselowitz · [01:12]

指向原始笔记的链接

但这里有一个隐藏假设:这套打法之所以有效,是因为你能承受「只错一瞬间」。
But there’s a hidden assumption here that it only works because you can afford to be wrong for an instant.
—— Jared Joselowitz · [04:47]

指向原始笔记的链接

靠盯仪表盘,仪表盘变红就意味着已经有一位真实患者受到了伤害。
By watching the dashboards, the dashboards are just going to, going red means that a patient was actually hurt.
—— Jared Joselowitz · [05:15]

指向原始笔记的链接

仿真是我们唯一能选择的、真正合乎伦理的选项。
Simulation is only the real ethical option we can go with.
—— Jared Joselowitz · [06:01]

指向原始笔记的链接

四组对话里有三组,多数人实际上认为模拟患者更逼真。
In three out of the four, the majority of people actually thought that the simulated patient was more realistic.
—— Jared Joselowitz · [08:51]

指向原始笔记的链接

结果显示,我们的裁判至少与真正的专家临床医生不相上下,甚至可能略好一些。
The results showed that our judge is at least on par, if not slightly better, than the real expert clinicians.
—— Jared Joselowitz · [10:22]

指向原始笔记的链接

你宁愿把不存在的危害误报出来,也不愿意漏报真正存在的危害。
You would rather over-call hazards that aren’t there than under-call hazards that are there.
—— Jared Joselowitz · [10:52]

指向原始笔记的链接

仅仅是格式上的变化,就曾被观察到能让基准成绩波动 76 个百分点;而重排少样本示例的顺序,能把模型在某些基准上从接近随机(约 50%)翻转到接近最先进水平。
Formatting changes alone have been seen to swing benchmark by 76 percentage points, and reordering few-shot examples flips a model from near-random, so near 50%, to near-state-of-the-art on some benchmarks.
—— Jared Joselowitz · [11:37]

指向原始笔记的链接

模拟是内循环——快、免费、可以在任何真人暴露前跑成千上万次;但真实患者是外循环,那才是唯一真正的证明所在。
Simulation is the inner loop. It’s fast. It’s free. You can do thousands of runs before anyone actually real is exposed. But real patients are the outer loop, and that’s where the only real proof is.
—— Jared Joselowitz · [15:40]

指向原始笔记的链接

所以模拟是必要的,但并不充分。
So simulation is necessary, but it’s not sufficient.
—— Jared Joselowitz · [15:51]

指向原始笔记的链接

重要的是,在试图通过监管时,你交付的不是模型,而是证据。
The important thing is that you don’t ship the model, you ship the evidence when trying to regulate.
—— Jared Joselowitz · [16:48]

指向原始笔记的链接

失败模式会变,框架不会。
The failure modes change. The framework doesn’t.
—— Jared Joselowitz · [18:15]

指向原始笔记的链接

② 出现在这些集

1 集

• 《给患者打电话之前,先让 AI 在仿真里跑几万遍:Euphonia 的医疗 AI 安全栈》 — 作为嘉宾

③ 他谈到的

点进去有真内容 —— 本页主要出口

Euphonia · DORA · 智能体 · Matrix · Patbot · BevJudge · 模拟 · 评审器 · 提示词优化器 · 成本矩阵

④ 也在聊「AI 安全」的人