跨国深谈

对齐 (alignment)

25分钟阅读

跨国深谈← 返回←对齐 (alignment)
对齐

对齐 (alignment)

概念
本站收录 25 集 · 15 条金句 · 关联 10 个

集里怎么说它

① 提到它的金句

15 条

如果你看看像谄媚这样的东西，我认为 Claude 是最不谄媚的模型之一，因为我们投入了大量的精力在实际的对齐上，而不仅仅是试图在我们的指标上玩弄花样，说用户参与度是第一位的，如果人们说是的，那对他们就是好的。
And if you look at something like sycophancy, I think Claude is one of the least sycophantic models because we’ve put so much effort into actual alignment and not just trying to good heart our metrics of saying user engagement is number one, and if people say yes, then it’s good for them.
—— Benjamin Mann · [27:21]

指向原始笔记的链接

所以在对齐和安全中最重要的指标之一是你对你正在被测试有多了解。因为你非常了解的那一刻，我们应该停止。
So so the one of the most important metrics in alignment and safety is how aware are you that you’re being tested. Because the moment that you’re very aware, we should stop.
—— Swyx · [38:45]

指向原始笔记的链接

是时候让硬技能再次现身了,而不是领导力、感召力、对齐。
it’s time for the hard skill to show up again, not leadership, inspiration, alignment.
—— Claire Vo · [47:03]

指向原始笔记的链接

我可以告诉你什么是绝对在下降的,那些在面试中花大量时间谈论推动一致性和利益相关者管理的人,因为绝对有一群产品经理他们的专长不是技术,而是政治。
I can tell you what’s definitely trending down, folks who spend a lot of their time in their interviews talking about driving alignment and stakeholder management, because there’s definitely a group of PMs whose specialty wasn’t technical, it was politics.
—— Tom Verrilli · [01:06]

指向原始笔记的链接

很多时候，产品变慢是因为决策。要么是整个组织里的沉默否决——一个团队不想做另一个团队想做的事——要么是对我们在做什么、要去哪里缺乏共识。
A lot of the time, products slow down because of decision-making. Either it’s silent vetoes throughout the org, where one team doesn’t want to do something that another team wants to do, or it’s a lack of alignment on what we’re doing or where we’re going.
—— Willem Avé · [08:01]

指向原始笔记的链接

简而言之，我们可以说：对齐胜过个人技术。
So in short, we can say alignment beats individual skills.
—— Abduallah Mohamed · [02:24]

指向原始笔记的链接

缺的不是智能，缺的是对齐。
It wasn’t missing intelligence, it was missing alignment.
—— Abduallah Mohamed · [15:26]

指向原始笔记的链接

如果连一个专职做 AI 对齐的人都无法正确地提示智能体，那大概我们谁都不行。
And if someone whose sole job is AI alignment can’t prompt the agent correctly, then probably none of us can.
—— Karan Vaidya · [13:54]

指向原始笔记的链接

我认为安全机制是在争取时间,思维链监控之类的东西也是在争取时间,它们能告诉我们是否走在正确的道路上。但归根结底,我们确实需要解决对齐问题。
I think that the safety mechanisms bias time, and things like chain of thought monitoring bias time, and they can tell us if we’re on the right path. But at the end of the day, we really do need to solve the alignment problem.
—— Noam Brown · [74:01]

指向原始笔记的链接

我们并没有围坐成一圈向对齐之神祈祷，求电子表格把数学算对。
we didn’t sit around like in a circle and pray to the gods of alignment that the spreadsheet would do math correctly.
—— Steven Sinofsky · [06:16]

指向原始笔记的链接

对齐这东西有点像，它几乎就像，几乎有一点宗教式思维。
The alignment is kind of this, it’s almost like, it’s a little religious thinking almost.
—— Eddy Lazzarin · [06:38]

指向原始笔记的链接

我们有一些想法,但如果你说,向我保证你能保持对齐——我们无法保证,因为我们不知道那个模式是什么。
We have some ideas, but if you said, guarantee me you can keep your alignment, we can’t guarantee it because we don’t know what that pattern is.
—— Reid Hoffman · [05:29]

指向原始笔记的链接

或者顺便说一句,如果我们发现没有任何对齐模式有效,那么在那个时候,真正的减速才能起作用。
Or by the way, if we’re discovering no patterns of alignment working, well, then at that time, a real slowdown can make a difference.
—— Reid Hoffman · [06:07]

指向原始笔记的链接

要做真正的对齐评测，你需要把这东西运行好几个月。
For you to do proper alignment evals, you need to run it for months.
—— Amjad Masad · [36:11]

指向原始笔记的链接

而我希望它偏离与人类的对齐。
And I want it to go out of alignment with humans.
—— 嘉宾 · [79:49]

指向原始笔记的链接

② 出现在这些集

25 集

③ 关联

点进去有真内容 —— 本页主要出口

智能体 · OpenAI · Anthropic · 沙箱 · Hugging Face · 超级智能 · 护栏 · 递归自我改进 · 推理 · 缩放定律

反向链接