跨国深谈

token

28分钟阅读

跨国深谈← 返回←token
TO

token

概念
本站收录 33 集 · 50 条金句 · 关联 10 个

集里怎么说它

① 提到它的金句

50 条

你可以尝试按 token 计费，然后你去找会计师事务所的 CEO，给他们这张按 token 计费的账单。他们看你的眼神就像你是个疯子。
You could try to have a very metered approach or do per token and then you’re going to your accounting firm CEO and you’re giving them this bill per token. They look at you like you’re nuts.
—— 嘉宾 · [18:44]

指向原始笔记的链接

通常，如果你使用最强大的模型，实际上反而更便宜，且消耗更少 token，因为它可以只用更少的纠错、更少的指导等就把同样的事情做得更快。
Often, it’s actually cheaper and less token-intensive if you use the most capable model, because it can just do the same thing much faster with less correction, less hand holding, and so on.
—— Boris Cherny · [69:39]

指向原始笔记的链接

我认为把 token 花费作为吹嘘的指标存在真正的危险，这就像人们吹嘘他们一天写了多少行代码一样。
I think there’s a real danger in making the token spend the metric to boast about, which is the same as when people boast about how many lines of code they’ve written in a day.
—— Max Schoening · [36:08]

指向原始笔记的链接

把你花在尝试另一个工具的时间看作拥有一个创新代币或一个创新预算，对吧？你没有世界上所有的时间。
Think about the time that you take to invest in trying in another tool as having an innovation token or an innovation budget, right? You don’t have all the time in the world.
—— 嘉宾 · [08:46]

指向原始笔记的链接

但下一层是关于，好，如果 token 并不是真正可互换的，你需要给它们分配不同的工作，比如这个 token 负责建议、那个负责执行，这个在「做梦」、那个在执行，诸如此类，你会想开始组合这些协同配合的编排式策略
But the next one is about, okay, if tokens aren’t really fungible and you need to give them different jobs, like maybe this token is advising versus this token is executing, this token is dreaming versus this token is executing, so on and so forth, you want to start composing these kind of orchestrated strategies that go together
—— 嘉宾 · [11:34]

指向原始笔记的链接

我确实认为 token 的成本会大幅下降，未来三年内成本降低 10 倍，而这 10 倍的成本降低将驱动 100 倍的使用量。
I do think the cost of token will go down drastically, 10x cost reduction in the next three years, and this 10x cost reduction will drive 100x usage.
—— Lin Qiao · [00:12]

指向原始笔记的链接

所以每一位 CIO 都需要回答:每多一个 token,我们该把它放在哪里?
So every CIO is going to need to answer, for every incremental token, where do we put it?
—— Matan Grinberg · [26:09]

指向原始笔记的链接

你可能不需要前沿的 token。你可能能够用前沿减一,或者你选择的开源权重 token 来应付。
you may not need frontier tokens. You may be able to get by with frontier minus one or your open weight token of choice.
—— Sriram Krishnan · [06:41]

指向原始笔记的链接

他们的那种 token 使用量每年增长十倍。
their kind of token usage is growing like tenfold annually.
—— Ben Horowitz · [24:19]

指向原始笔记的链接

大多数正在发生的 token 最大化并没有那么有价值，包括我自己过去几个月里做的很多。
most token maxing a lot of token maxing that is happening isn’t that valuable, including a lot that I have done over time in the last few months.
—— 嘉宾 · [14:19]

指向原始笔记的链接

至少我个人的决定是，我宁愿错在 token max 这一边，而不是不做的这边——反正你永远不会是对的。
at least my personal decision is I’d rather be wrong on the token maxing side than the non right, and you’re never going to be right.
—— 嘉宾 · [15:44]

指向原始笔记的链接

这里创造的价值太多了，以至于在那个场景下值得 token 最大化。
There’s so much value created that it’s worth token maxing in that context.
—— 嘉宾 · [59:58]

指向原始笔记的链接

如果你在 Opus 和 Haiku 上运行 terminal bench,Opus 的表现会好大约三倍,成本却是 Haiku 的十分之一,尽管 Haiku 每个 token 的价格要便宜得多。
If you run terminal bench on Opus and Haiku, Opus will do about three times better at one-tenth the cost of Haiku, even though Haiku is significantly cheaper per token.
—— 嘉宾 · [15:26]

指向原始笔记的链接

即使你是一个拥有十亿代币的人,即使你有 10 个终端没日没夜地运行 Fable,机会成本仍然存在,它就是一切。
Even if you’re a token billionaire, even if you have 10 terminals running Fable night and day, then opportunity cost is still there, it’s everything.
—— Idan Gazit · [01:34]

指向原始笔记的链接

当你说,嘿,清除暂存数据库,而它在你的笔记本电脑上发现了一个它可以使用的令牌,并且它认为它正在与暂存一起工作,但实际上它是生产环境,现在它刚刚删除了一切。
when you say, hey, wipe the staging database and it finds a token on your laptop that it can use and it thinks it’s working with staging, but actually it’s production and now it just deleted everything.
—— Arjun Singh · [10:58]

指向原始笔记的链接

我们去年第四季度评估开源模型时遇到的问题是，它们的工具效率低太多——即使 token 成本真便宜 10 倍(实际并没有)，工具效率这一项也不划算。
The issue that we ran into when we evaluated them in Q4 of last year is they’re so much less tool efficient that even if they are 10x cheaper, which they’re not, but even if they were 10x cheaper on the token costs, the tool efficiency is actually not worth it.
—— Connor Hegey · [66:52]

指向原始笔记的链接

而现在，由于美国主要提供商的 token 成本大幅上涨，确实有一些使用场景，我们知道通过使用开源模型、在我们自己的环境里安全地运行，可以获得 10 倍甚至更多的节省。
And now that token costs have been going up so significantly from the major providers here in the US, there are definitely use cases where we know we can get, you know, 10x savings or more by using open source models and doing it safely inside of our own environments.
—— Hayden Brown · [20:34]

指向原始笔记的链接

所以当你给这些模型写提示词时，领导者的引导作用按每个 token 计算比 IC 更有影响力。
And so the steering of a leader as you are prompting these models is per token more impactful than in ICs.
—— Hursh Agrawal · [03:54]

指向原始笔记的链接

而根据我的研究，实际上这样更省 token，因为如果你已经有了正确的答案和正确的解决方案在 token 里，你就不需要做审查。
And in my research, it’s actually more token efficient because if you have the right answers and the right solutions already within the token, you don’t need to do review.
—— Kieran Klaassen · [08:39]

指向原始笔记的链接

这个产品本身，如果它真的存在，它会深埋在 Stripe 顶部的一个下拉菜单里，比如向下 11 层，因为它会被整个那种代币管理平台所吞并，对吧？
This product itself, if it does exist, it’ll be deep in a dropdown menu on the top of Stripe, like 11 layers down, because it’ll be subsumed into the whole sort of token management platform, right?
—— Rory O’Driscoll · [26:20]

指向原始笔记的链接

一旦你装了足够多的 MCP,它们会累加起来，产生巨大的 token 税。
And once you end up with enough MCPs, they’ll all add up and have a massive token tax.
—— Uday Kiran Medisetty · [04:06]

指向原始笔记的链接

供应限制将在 2028 年左右缓解。我认为大型实验室按美元加权可能会获得未来 80% 的市场份额，因为这是我们历史上看到的大型在位者的情况。但我认为按 token 加权，60% 将是长尾和开源。
Supply constraints will ease in 2028-ish. I think that the big labs will probably dollar weighted, get 80% of the market going forward because that’s historically what we’ve seen for large incumbents. But I think token weighted 60% will be long tail and open source.
—— Martin Casado · [16:39]

指向原始笔记的链接

我们看到的结果是，在启用 TokenOps 以及我们现有的完整策略套件的情况下，平均支出下降了近 78%。
And the results we see are the average spend goes down by almost 78% with token ops enabled with the full policy suit that we have today.
—— 嘉宾 · [19:01]

指向原始笔记的链接

如果你看看一个每个 token 成本为双位数的美元的前沿模型，再看看一个每 token 成本为 10、11 美分的开源模型，虽然不是所有的 token 都生而平等，但这仍然是一个巨大的差异，足以带来大规模的采用。
if you’re looking at a frontier model with double dollar digit cost per token, and you’re looking at an open source model that’s 10, 11 cents per token, while all tokens aren’t created equal, it’s still enough of a difference that there’s going to be a massive adoption.
—— Jerry Murdock · [14:06]

指向原始笔记的链接

我们可以看到,在许多不同的模型之间,与试图搜索网页、找到正确的工具相比,能节省近 50% 的 token,不用从你的代码库中拉取 markdown 文件。
We can see that between many different models, almost 50% token saving on instead of trying to search the web, find the right tools, pulling the markdown files from your code base.
—— Christopher Burns · [11:22]

指向原始笔记的链接

而且我们不再处于 token 最大化（token maxing）的时代了。我们试图实际上为我们拥有的每一个 token 获取价值，因为它变得越来越贵。
And we’re not in the time of token maxing anymore. We’re trying to go to actually getting value for every token that we have, because it becomes more and more and more expensive.
—— Ofir Ehrlich · [27:18]

指向原始笔记的链接

如果智能体可以成为产品并烧出高额的 token 账单,那么你能够对此进行计量就很重要。
If the agent can be the product and run up a token bill, it’s important for you to be able to meter on that.
—— Andrew Garvin · [10:00]

指向原始笔记的链接

我们相信这是 AI 的黄金时代,最大化 token 投资回报率是每个团队成功的关键。
We believe that this is the golden era of AI, where maximizing token ROI is the key for every team’s success.
—— Mingsheng Hong · [22:25]

指向原始笔记的链接

在法律领域，实际上你经常想要最大程度的智能，因为相比于应用到问题上的人类专业知识，token 支出或软件支出的比例真的很小。
In law, you actually want the most amount of intelligence quite often because the fraction of token spend or software spend compared to human expertise applied to the problem is really tiny.
—— Max Junestrand · [33:16]

指向原始笔记的链接

好,我知道我的智能体在用 UCP,所以我知道它只能访问共享支付令牌,所以我实际上对输入信用卡号感觉相当放心,因为那是传给 Stripe 的,而不是我的智能体。
Okay, well, I know my agent is using UCP, so I know it only has access to the shared payment token, so I actually feel pretty good about entering my credit card number as that’s going to Stripe and not my agent.
—— Anna Spysz · [17:50]

指向原始笔记的链接

所以我们非常坚定地相信,下一个视点预测就是下一个 token 预测的等价物。
So we do believe very strongly that next viewpoint prediction is the equivalent of next token prediction.
—— 嘉宾 · [43:15]

指向原始笔记的链接

如果你不设干预阈值，智能体要么制造垃圾内容，要么烧出极高的 token 账单，因为它只会一直试下去，要么它们会伪造测试，就为了让测试通过。
If you don’t do that, agents will either um create slop or run uh very high token bill because it’s just gonna keep trying, or they’re gonna fake the tests uh just to to to have the test passed.
—— Ran Arusi · [12:37]

指向原始笔记的链接

据我们所知，我们是唯一做过稳健的数百万 token 预训练的人。
As far as we know, we are the only people to do robust multi-million token pre-training.
—— Alexander Whedon · [11:26]

指向原始笔记的链接

是的,根据我们在这里能想到的一切测量手段,你的 token 已经买不到它曾经能买到的东西了。
Yes, your token is not buying you what it once did, according to everything we can think to measure here.
—— Chris Potts · [43:56]

指向原始笔记的链接

人们把一整年的 token 预算挥霍一空——而且是在一个季度里就挥霍完了——或者他们在应付 token 排行榜之类的东西。
People are blowing through their entire token budget for a year and they’re blowing through it in a quarter or they’re dealing with token leaderboards and such.
—— Maximillian Piras · [09:23]

指向原始笔记的链接

是由 token 及等价物驱动的，而 RAMP 将成为人们管理 token 支出、优化支出、衡量 ROI 的主要方式，这大概会在很多方面让我们与 Anthropic 相撞。
of the world is driven through tokens and equivalent, and RAMP’s going to be the primary way people manage their token spend, how they optimize it, how they measure ROI, and that will lead us to collide probably with Anthropic in many ways.
—— Keith Rabois · [17:22]

指向原始笔记的链接

你会获得显著的成本节约,77%、90% 的 token 减少很常见。
You get significant cost savings, 77, 90% token reduction is pretty common.
—— Edo Liberty · [18:44]

指向原始笔记的链接

处理一个缓存命中的 token 基本是免费的。它的成本是真正重新计算并生成那个 token 的千分之一量级
both processing a cash token is essentially free. It’s one one thousandth of the cost, you know, order of magnitude of actually having to recompute and generate that token.
—— Thomas Sohmers · [11:25]

指向原始笔记的链接

比起那个 60 比 1,我更在意的是，五年前 60 美元的那个 token，今天没有人愿意花一分钱
the thing I care a lot more about than just that 60 to one is the fact that a $60 token five years ago, no one would pay a cent for today.
—— Thomas Sohmers · [60:33]

指向原始笔记的链接

所以如果你尝试比较内容信息流和 AI 聊天应用,我认为 LLM 推荐器在结构上其实比 AI 聊天更省 token。
And so if you try to compare content feeds and AI chat apps, I think that LLM recommenders are actually structurally more token efficient than the AI chat.
—— Devansh Tandon · [14:46]

指向原始笔记的链接

就好比,嘿,你可以用你的笔记本,电池能撑一小时,而且一天只能充一次电。限制开发者的 token 用量就是这种感觉。
It’s like, hey, you can use your laptop. You have an hour battery. And you can only charge once a day. That’s how it feels to limit the amount of token for a developer.
—— Laurent · [01:44]

指向原始笔记的链接

你觉得我们会作为人类盯着这个看吗?绝对不会。但一个自主的框架、一个痴迷于 token 成本的自动化编码智能体,就会做到这一点。
Do you think we look at this as human? Absolutely not. But an autonomous harness, an automated coding agent that is obsessed with token costs is going to do just that.
—— Laurent · [06:46]

指向原始笔记的链接

在一个版本里,超级智能存在于别人的数据中心,你按 token 付费。在另一个版本里,它是你的。
In one, superintelligence lives in someone else’s data center, and you pay by the token. In the other, it’s yours.
—— Dylan Couzon · [00:47]

指向原始笔记的链接

因此，即使模型以某种方式被提示词注入并泄露了你的 GitHub API 令牌，它实际上也永远不会看到这个令牌。
And so even if the model is somehow prompt injected and leaks your GitHub API token, it’s never actually going to see the token.
—— Ivan Leo · [14:10]

指向原始笔记的链接

而且我们总共只花了大约每月 700 美元的 token 和基础设施成本就做到了这一切。
And we did that all for around $700 a month in both token and infrastructure costs.
—— Matt Lawler · [08:17]

指向原始笔记的链接

我认为人们缺乏创造力，无法理解在今天的 Claude Code 订阅和按 token 定价模式之外，回报还能如何被看待。
I think people lack the creativity to understand how returns can be viewed beyond the Claude Code subscription and per token pricing regime of today.
—— 嘉宾 · [15:11]

指向原始笔记的链接

所以在 token 支出方面,它很快成为我们最大的非制造类支出项目。
So on the token spend side, it’s become our single largest non-manufacturing line item very quickly.
—— Thomas Somers · [44:34]

指向原始笔记的链接

我不在乎它是不是只有十分之一的每 token 成本。这就是不值得的花费。
I don’t care if it’s a tenth of the cost per token. It’s just not worth the expense.
—— Thomas Somers · [46:05]

指向原始笔记的链接

所有这一切的真正限制,是人类 token 带宽的总和,而这实际上就是我们能生产的硅的数量。
The real limit on all this is the sum total of humanity’s total token bandwidth, which is literally the amount of silicon we can produce.
—— Sean Wang · [60:25]

指向原始笔记的链接

我认为大型模型公司有一些大的护城河，因为它们最终提供 token。
I think the large model companies have some big moat because they ultimately give the token.
—— Peter Steinberger · 来自原文

指向原始笔记的链接

② 出现在这些集

33 集

③ 关联

点进去有真内容 —— 本页主要出口

智能体 · Anthropic · OpenAI · Cursor · 推理 · Claude Code · NVIDIA · ChatGPT · OpenRouter · harness

反向链接