Gabe
① 他说过的话
11 条
我们当时有一个非常强烈的直觉，就是产品将会是模型。
指向原始笔记的链接
The really strong intuition that we had at the time was the product is going to be the model.
—— Gabe · [00:00]
我认为当时 GPT-4 已经足够好了，即使模型不再继续进步，也有很多事情可以做。
指向原始笔记的链接
I think already GPT-4 was so good that it was clear that even if the models didn’t continue improving, there was so much you could do.
—— Gabe · [06:19]
而我认为反驳是，你本可以用个人电脑做出同样的论点，而实际上计费反而上升了，因为工作变得更复杂了。
指向原始笔记的链接
And I think the counter is like you could have made the same argument with the PC and it’s like actually billables went up because the work got more complex.
—— Gabe · [16:59]
我不需要 ROI 指标就知道这从根本上改变了我的方式。
指向原始笔记的链接
I don’t need ROI metrics to know that this like fundamentally changes the way that I like.
—— Gabe · [19:44]
我认为真正有价值的数据——现在人们正在意识到它有价值——是 RLHF 数据或者推理数据。
指向原始笔记的链接
The data that I think is really valuable that I think now people are realizing is valuable is the RLHF data or the reasoning data.
—— Gabe · [23:17]
特权、保密性——既有技术层面的安全，也有法律层面的安全。比如，如果我训练了一个模型，我是否就放弃了特权保护？我认为还没有任何人知道答案。
指向原始笔记的链接
Privilege, confidentiality, like both security on the technical side and security on the legal side of I don’t think anyone knows the answer yet of if I train a model, do I waive privilege?
—— Gabe · [25:20]
我们的总体感受是，我们永远不会去预训练模型。
指向原始笔记的链接
Our general sense is we will never pre-train models.
—— Gabe · [33:33]
但我会说，其实具体的工具并不是那么重要，真正重要的是思考产品的形态如何与模型契合。
指向原始笔记的链接
But I would say that was actually less of like the specific tooling wasn’t as much of what mattered, but it was really thinking about how does the shape of the product fit with the models.
—— Gabe · [39:07]
对我来说，这很多就像是一个自动驾驶的迷你版本，就是，最终你在大规模上证明这件事的唯一方式，是让这个东西谈判足够多的 NDA，以至于人们就是信任它。
指向原始笔记的链接
Like to me, a lot of this is like, this is like a mini version of self-driving where it’s like, the only way eventually you prove this at scale is you have this thing negotiate so many NDAs that people just trust it.
—— Gabe · [42:18]
人数增长来扩展营收——转变为受分发约束，也就是在五年内他们基本上能把自己的全部专业知识放进这个系统里，这个系统可以完成客户事务的大部分工作，然后想出办法把它卖给客户。
指向原始笔记的链接
Headcount to distribution constrained where they are able in five years to basically take all of their expertise, put it into this system that can do large parts of a client matter and then figure out ways to sell that to customers.
—— Gabe · [43:39]
所以我认为我们在五年内希望看到的是，那样的工作是由一个由律所和那个客户训练的模型来完成的。
指向原始笔记的链接
And so I think the thing we see, we would want to see in five years is like that work is done by a model trained by a law firm and that client.
—— Gabe · [44:22]
② 出现在这些集
2 集
- 《Harvey 联合创始人 Gabe：产品就是模型，AI 律所的五年赌注》 — 作为主持
- 《经验差距：让智能体越用越聪明》 — 作为主持
③ 他谈到的
点进去有真内容 —— 本页主要出口
Harvey · 智能体 · 评估 · 后训练 · OpenAI · Arjun Karanam · 工作流 · 轨迹数据 · RL 环境 · 持续学习