AI Lurches Toward Recursive Self-Improvement — 文稿

M10:00

嗨，朋友们。

Hi, friends.

M10:38

大家脑海中都萦绕着三个字母：R、S、I。

The 3 letters on everyone's minds are R, S, and I.

MIT的一篇新论文展示了更大模型如何编辑更小模型的实际权重，这让大家开始思考。

A new MIT paper showing how a bigger model can edit the actual weights of a smaller model has people thinking about it.

而试图理解下一轮增长来自何处的Wall Street投资者，也在深入挖掘RSI可能意味着什么。

And Wall Street investors who are trying to understand where the next leg of growth comes from are also digging deep into what RSI might mean.

在本集中，我们将探讨一组故事，它们的主线是RSI，以及更广泛地，AI接下来会怎样。

On this episode, we're looking at a set of stories whose throughline is RSI, and more broadly, what comes next for AI.

但这一切要从Elon Musk对整个个人智能体领域提出质疑开始。

But it starts with Elon Musk casting doubt on the entire personal agent space.

The AI Daily Brief是一档每日播客和视频节目，内容涉及AI领域最重要的新闻和讨论。

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好了，朋友们，在开始之前先快速说几个通知。

All right, friends, quick announcements before we dive in.

M11:33

还有两个快速通知。

2 More quick announcements.

M11:56

第二条通知是，我这周在旅行庆祝周年纪念，基本上是丛林中向你们直播。

The second note is that I am traveling this week for my anniversary and basically coming to you live from the jungle.

所以如果你们碰巧听到鸟叫声或流水声，或者说比平时更多自然的声音，原因就在这里。

So if you happen to hear bird sounds or water trickling, or let's just call it more nature than usual, that is why.

我希望你们在接下来几天能享受这些丰富的自然音。

I hope you enjoy these various nature sounds for the next couple of days.

下周我们会恢复正常。

Next week, we will be back to normal.

整个2026年，尤其是过去几个月，一直在热议的一个话题是，AI模型越来越接近能够自我改进，这对AI行业乃至整个社会意味着什么。

One of the conversations that has been swirling around quite a bit Throughout 2026, but especially over the last couple of months, is what it means for the AI industry and society more broadly that AI models are getting closer to being able to improve themselves.

在Alphabet最近的财报电话会议上，CEO Sundar Pichai表示，我们正处于AI帮助我们提升训练和发布速度的阶段。我预计未来12个月会因RSI而持续加速。

On Alphabet's recent earnings call, CEO Sundar Pichai said, we're at the point where AI is helping us increase our training and launch velocity.

不过，今天的故事并不全是关于RSI的。

I would expect the next 12 months to see continued acceleration because of RSI.

但有一条主线贯穿其中：AI的持续加速与AI的采用如何交汇，这如何影响市场潜力，以及RSI的实际状况。

Now, today's stories are not exactly all about RSI.

我们三则故事中的第一个，源于Elon Musk一个看似随意但实则一点也不随意的帖子，他在周末表示，运行GrokBot的成本与收入相比高得难以持续。

But there is a through line in how the continued acceleration of AI meets adoption of AI, how that impacts market potentials, and what the state of RSI actually is.

这件事的起因是，上周末GrokBot用户抱怨他们开始遇到之前从未有过的使用限制。

The first then of our troika of stories comes after a seemingly nonchalant, but not really at all nonchalant post from Elon Musk, where he said over the weekend, the cost of serving GrokBot was unsustainably high relative to revenue.

Elon是在回应某人要求提高或重置这些使用限制的请求。

The genesis of this had been that over the weekend, GrokBot users complained that they were starting to hit usage limits that had never been a problem before.

不过，Elon并没有给出更多细节。

Elon was responding to someone's request to increase or reset those usage limits.

不清楚用户是通过哪个付费计划访问GrokBot的，但对于密切关注AI领域的人，特别是市场悲观者来说，这个帖子犹如掷下一枚重磅炸弹。

Now, Elon didn't give any more details.

对某些人而言，Elon过去几周的言论中存在一种奇怪的矛盾。

It's unclear through which plans users were accessing GrokBot, but for those watching the AI space closely, especially market bears, the post hit with the force of a bomb.

就在几天前，他刚发帖说，随着超级智能富足时代的到来，为遥远未来存钱是没有意义的。

For some, there was a weird contradiction in Elon's communications over the past couple of weeks.

但正如Bruno Lemos等人所指出的，仅仅几天后，他又在说成本太高了。

Just a couple of days ago, he had posted, saving money for the distant future is pointless with the advent of superintelligent abundance.

Benjamin DeCracker追溯到更早，重新分享了几周前Elon的一个帖子，当时他说，从未见过任何东西像GrokBot增长这么快，而且没有免费层级，这意味着人们获得的实际价值超过了他们支付的费用。

But as Bruno Lemos and others pointed out, just a couple days later, he's talking about costs being too high.

DeCracker写道，两周前，GrokBot还是一个令人难以置信的成功故事，推动着xAI的收入增长。

Benjamin DeCracker went a little bit farther back, resharing an Elon post from a couple of weeks ago when he said, never seen anything grow as fast as GrokBot, and there is no free tier, which means people are deriving real value in excess of what they pay.

现在却变得不可持续而且在亏损？

DeCracker writes, 2 weeks ago, GrokBot was an incredible success story powering xAI's revenue growth.

对某些人来说，Elon奇怪的沟通完全比不上对收费计划那种被背叛的感觉。

Now it's unsustainable and loses money?

对一些人来说，埃隆那奇怪的言论完全比不上对套餐的背叛感来得重要。

For some, Elon's weird communications were entirely secondary to the feeling of betrayal around the plans.

用户Shrek说，这不是我们使用不够或者选错计划的问题。

User Shrek said, This isn't about us not having enough usage or picking the wrong plan.

我一直在你们这里买12英寸的披萨，你们之前一直给我正常的12英寸的。

I've been buying the 12-inch pizza from your place forever, and you guys have always given me a normal 12-inch.

然后有一天我打开盒子，结果是个6英寸的。

Then one day I open the box and it's a 6-inch.

你跟我说成本太高，所以以后12英寸其实就是6英寸，如果我还没吃饱，那是我没选对尺寸的错。

You tell me costs are too high, so from now on the 12-inch is actually 6 inches, and if I'm still hungry, it's my fault for not choosing the right size.

这也太离谱了。

That's insane.

如果你们需要降低成本，提前告诉我们就行了。

If you needed to cut costs, just tell us ahead of time.

如果你们的披萨好吃，我还想要的话，我大可以多下单两三个这种假12英寸的。

If your pizza is good and I still want it, I can just order 2 or 3 of these fake 12-inch ones.

或者如果我没吃饱，我也可以去别的地方。

Or if I'm not full, I can go somewhere else.

不管哪种情况，那都是我的决定，不是你的。

Either way, that's my call, not yours.

但你不能一直卖给我正常的12英寸，然后突然缩小成6英寸，连个通知都没有，甚至连让我调整的预警都没有。

But you don't get to keep selling me a normal 12-inch and then suddenly shrink it to 6 inches with zero notice, not even a heads-up so that I can adjust.

别反过来说我选错了订阅套餐。

Don't turn around and say I picked the wrong subscription.

问题是你们单方面改变了交付的内容，却谁都没通知。

The problem is you unilaterally changed what you were delivering without telling anyone.

其他人认为这不可避免地是其他个人智能体的好消息。

Others thought that inevitably this was good news for the other personal agents.

Alex Michael写道，这对Dot和Duet来说是个巨大的胜利，很可能会把大量用户直接推回给他们。

Alex Michael wrote, this is a huge win for Dot and Duet and will most likely drive a lot of users right back to them.

Nous Research，也就是Hermes的开发者，当然对此大加利用。

Nous Research, the makers of Hermes, of course jumped all over it.

他们还分享了关于GrokBot用户如何迁移到Hermes的具体指导。

And shared specific instructions on how GrokBot users can migrate to Hermes.

但最有趣的讨论是，Elon和xAI遇到的问题，与其说与xAI以及他们在AI竞赛中的地位有关，不如说更多地关乎这类产品的基本面。

But the most interesting discussion, whether the issues that Elon and xAI are running into are less about anything having to do with xAI and their position in the AI race and more about the fundamentals of this product category.

来自OpenCode的Dax转发了Elon并说，这是我们对这类产品形态不那么兴奋的另一个主要原因，这里他大概是指个人智能体。

Dax from OpenCode reposted Elon and said, this is another major reason why we weren't super excited about this shape of product, by which presumably Dax means personal agents.

他接着说，人们不愿意为此付费，所以它必须是一个漏斗顶端的、能导向其他东西的产品。

Continuing, he says, people don't want to pay for it, so it has to be a top-of-funnel product that leads to something else.

还不确定那个其他东西会是什么。

Not sure what the something else could be yet.

现在，细心的观察者会注意到，这种关于免费消费者层级上永远不付费的免费AI用户构成的成本中心，以及它如何影响公司更广泛变现能力的讨论，已经存在很久了。

Now, careful observers will note that this sort of conversation about the cost center that is free AI users on free consumer tiers that will never pay and how it impacts the ability of the company to monetize more broadly has been around for a very long time.

在2025年秋季的Bubble Debates中，这是一个主要讨论话题，当时人们的基本担忧是，OpenAI的ChatGPT异军突起的成功实际上可能成为一种阻碍，因为支持所有这些免费用户需要耗费巨大成本。

It was a major subject of discussion in the fall 2025 vintage of Bubble Debates, where people were basically concerned that OpenAI's breakaway ChatGPT success could actually be a hindrance because of how much it costs to support all of those free users.

自从我们开始谈论个人智能体，特别是今年它们变得更真实以来，就一直有人应用同样的论断：既然消费者不愿为这类产品付费，它们就只会成为提供者的一种成本中心，必须通过其他方式来弥补。

For as long as we have been talking about personal agents, especially this year as they've gotten more real, there have been some who have applied those same sort of claims that since consumers aren't willing to pay for this category of product, they're just going to be a cost center for the people offering them, and they're going to have to be made up for in other ways.

而这仍然是一个悬而未决的争论。

And that remains an open debate.

当然，市场怀疑论者迅速开始探讨这对所有提供这类产品的其他公司可能意味着什么，其中非常引人注目的是Meta，Meta一直因其在这一类别中的领先地位而在更广泛的AI竞赛中占据了一席之地。

And certainly market skeptics jumped right to what this might mean for all the other companies offering these products, which includes very notably Meta, who have been riding high on the fact that Meta's leading place in this category has given them a foothold in the AI race more broadly.

当然，悲观主义者们很高兴将此视为既成事实，认为泡沫已破，游戏结束。

Certainly the bears were very happy to jump all over this as fait accompli that the bubble was done and the game was over.

Eric Sky写道，Elon将会退出这场竞赛。

Eric Sky writes, Elon is going to abandon the race.

第二次警告。

Second warning.

Nick Huber，他一向专注于传统又无聊的生意，对很多类型的科技初创公司、当然也包括AI，都抱有怀疑，他傲慢地表示，这正是现在AI的现状。

Nick Huber, who is all about traditional boring businesses, and very skeptical of many categories of tech startups and certainly AI, got on his high horse and said, this is all of AI right now.

音乐终会停止。

The music will stop.

最好找到你的座位。

Better find your chair.

风投资金用完了，意味着价格大幅上涨，意味着用户流失，意味着估值暴跌。

VC money runs out, means prices rise dramatically, means users drop off, means valuations plummet.

Hedgeye Markets认为这只是补贴时代结束的又一个信号。

Hedgeye Markets thinks it's just another sign of the end of the subsidy era.

他们写道，Grok用户本周初就用完了使用天数，马斯克直接告诉了他们原因。

They wrote, Grok users ran out of usage days early this week, and Musk was straight with them about why.

这就是我在本周通讯中写到的补贴的终结。

This is the end of the subsidy I wrote about in this week's newsletter.

AI公司把套餐定价低于成本来吸引用户，而这只有在投资者补贴差价的情况下才可行。

AI companies price their plans below cost to win users, and that only works while investors cover the difference.

根据今年夏天的一项分析，一个200美元的ChatGPT Pro套餐，按常规API价格计算，一个月的实际成本大约是一万四千美元。

A full month on a $200 ChatGPT Pro plan would cost about $14,000 at regular API prices, according to one analysis this summer.

我觉得他指的是SemiAnalysis，但他提到这是今年夏天的讨论，这一点值得花点时间聊聊。

Now, I think by that he's referring to SemiAnalysis, but his note that this was the discussion this summer is worth taking a moment on.

你可能还记得，我们在四月底做过一期关于AI补贴时代的节目。

You might remember that we did an AI subsidy era episode back at the end of April.

那期节目不仅论证了AI补贴时代即将结束，还试图为企业提供一套适应这种变化的行动方案。

The episode not only argued that the AI subsidy era was coming to an end, but tried to provide a set of actions for businesses to adapt to it.

所以毫无疑问，这不是什么新话题，而且它早就不可避免了。

So make no doubt that this is not some new conversation, and it has been inevitable for quite some time.

出于某种原因，我们夏天那几个月被其他事情分散了注意力，就没再讨论这个了。

For whatever reason, we got distracted by other things for a couple months over the summer and stopped talking about this.

但现在，泡沫分析再次铺天盖地，再加上Anthropic接近IPO带来的批判性分析，这个话题又回来了，因为关于其影响的争论显然从未解决过。

But now that there's so much bubble analysis once again, and critical analysis with Anthropic getting close to IPO, the story is back because of course the debate about the implications was never resolved.

这些补贴结束后的看空观点。

The bear case with these subsidies ending.

基本就是说，当AI公司真正需要为产品收取正确或自然的价格时——也就是向客户收取token的实际成本，再加一点自己的利润，而不是由风投和私募股权投资者补贴的版本——人们并不认为或者负担不起这个自然价格下的智能价值，从而无法像以前那样继续消费。

Is basically that when AI companies actually have to charge the correct or natural rate for their product, i.e., charge customers what the tokens actually cost, plus a little bit of margin for themselves, not a version of that that is subsidized by someone else, like venture capitalists and private equity investors, that people don't value or can't afford to value the intelligence at that natural price enough to continue spending in the way that they have been.

那就意味着所有这些公司的估值都不对了。

And thus it means that all the valuations of all these companies are off.

这是一个可能的结果，我认为市场参与者试图弄清楚如何将这种风险定价是完全合理的。

That is a possible outcome, and I think it's completely reasonable for market actors to try to figure out how to price in that risk.

事实上，总的来说，我想说，虽然我绝不是AI看空者，但我对目前市场上的AI看空者完全没有意见。

In fact, in general, I will say here that despite the fact that I am very much not an AI bear, I have no issue at all with the AI bears in the market right now.

我认为如果所有人都过于迷恋Anthropic和OpenAI，以至于不去问这些关键问题，那才会更危险。

I think it would be much more dangerous if everyone was too enamored of Anthropic and OpenAI to actually ask these critical questions.

我认为泡沫的可能性在某种程度上与看空者的声音大小及其信息在广大投资者中的共鸣程度成反比。

I think that likelihoods of a bubble are decreased in proportion in some ways to the loudness of the bears and the resonance of their message with a broad base of investors.

当然，我也认为看空分析的质量参差不齐，听那些多年来一直判断错误、不断转移目标以求正确的永久看空者的话，可能并非最佳选择。

Now, I also, of course, think that there's a wide spectrum of the quality of bearish analysis, and that perhaps listening to the perma-bears who have gotten it wrong for years and continue to move the goalposts to get it right might not be your best play.

但总体来说，我认为他们的声音是市场情绪的重要组成部分，而且投资者花时间将他们讨论的风险纳入定价，这是有价值的。

But overall, I think their voices are an important part of the market sentiment, and I think it's valuable that investors are taking time to price in the risks that they're talking about.

所以，话虽如此，我并不特别认为转向对所有用户按API定价会带来多大改变。

So all that said, I don't particularly think that the shift to API pricing for everyone actually changes all that much.

基本上，我认为套餐用户分为两类。

Basically, I think that the plan users split into one of two categories.

一类人觉得那些带有更多效率权衡的模型对他们的使用场景来说已经足够好了，他们继续使用每月20美元或30美元的套餐，因为对他们的用例来说，变化不大。

The people for whom the models that come with more efficiency trade-offs are good enough for their types of uses, who stay on their $20 or $30 a month plan because for their use cases it doesn't change all that much.

而另一类人则从他们当前使用的套餐（无论是20美元还是200美元）中获得了足够价值，以至于他们愿意真正为自己消耗的智能付费，并把它当作一项业务成本。

And then on the other side, the people who are getting enough value out of whatever plan they're on, whether it's $20 or $200, that they're willing to actually pay for the intelligence they consume and just treat it as a cost of doing business.

我当然不是一个非常典型的例子，但除了我的高级套餐，我还在很多地方运行着API，而且毫无疑问，我从它们那里获得的价值完全值得这个成本。

I'm certainly not a super normal example of this, but I have APIs running all over the place in addition to my premium plans, and there's absolutely no doubt that the value that I get out of them is well worth the cost.

我用那些API做事，是的，成本不低，但这个成本仍然远低于同等人力劳动的成本。

I do things with those APIs for, yes, a premium cost, but a premium cost that's still far below what the comparative human labor would have cost.

顺便说一句，这不是一对一的替代，因为人力劳动在金钱和时间上（招聘和管理这些人的时间）成本都太高了。

Which, by the way, is not a one-to-one replacement because the human labor would have cost too much in both actual money as well as time to recruit and manage those people.

所以从经济角度看，这是我以前不会花的、额外的新开支。

So from an economic perspective, this is net new money that I'm spending that I would not have spent before.

我认为我们在2026年看到的，基本上自从智能体时代兴起以来，从实际百分比来看，很大一部分收入来自那些已经通过API付费的重度用户。

What I believe that we've seen in 2026, basically since the rise of the agentic era, is that a huge amount of the revenue on an actual percentage basis is coming from power users who are already paying via the API.

事实上，我认为有两个看空观点并列在一起，形成了一种奇怪的对比。

In fact, I think there's a bit of a weird contrast between two bear arguments sitting next to one another.

是不是说，当人们开始通过API付费时，他们会流失，然后停止使用AI，因为他们觉得从中得不到足够的价值了？

Is it that when people start paying on the API, people are going to drop off and stop using AI because they don't get enough value out of it anymore?

还是说市场过度集中在极少数重度用户身上？

Or is the market too concentrated among a very small number of power users?

当然，这里更微妙的中间观点，就是像Chamath Palihapitiya这样的人提出的，他们认为目前有太多重度用户是靠风投资金补贴的，这可能会带来一些严重的问题。

Now, of course, the more nuanced in-between argument here is the one that you've seen people like Chamath Palihapitiya make that argues that too many of those power users are currently subsidized by VC dollars, and that could present some significant problems.

我认为这是一个更有意思的探讨方向。

I think that's a much more interesting thing to explore.

不过目前我们还不知道，我们所有的只是这种解读。

Still, right now, we don't know, and all that we have is this sort of interpretation.

大家好。

Hello everyone.

M114:26

当然，Elon Musk确实会毫不避讳地在市场上煽风点火，助长恐慌。

Certainly it's the case that Elon Musk is not above fanning the flames of panic in the market.

上个月CoreWeave的债券表现特别糟糕，Musk评论说‘哎哟’，这句话被金融媒体广泛报道。

After CoreWeave's bonds had a particularly bad day last month, Musk commented, yikes, which got picked up across the financial press.

后来在十月，他说，对于任何没有独立于AI数据中心之外的大额收入的公司，债务市场几乎已经彻底关闭了融资渠道。

And later in October, he said, debt markets have almost entirely shut off access to money for any company that doesn't have substantial revenue independent of AI data centers.

有不少人认为这有点像是某种高维棋局，Musk在推动市场进一步抛售，以期进行低价收购或打击竞争对手。

There are more than a few people who think this is some version of 4D chess, with Musk pushing the markets to sell off harder in hopes of making some cheap acquisitions or harming rivals.

但目前我们就是不知道。

But right now, we just don't know.

我们确定的是，仍然有一些公司在个人智能体上下重注，尽管Elon似乎在退缩。

What we do know is that there are still some companies who are betting very heavily on personal agents, even if Elon seems to be pulling back.

a16z的Olivia Moore认为，谈到Meta的Muse，真正的重点其实不在消费者端，而是面向小型企业的Muse。

Olivia Moore from a16z thinks that when it comes to Meta's Muse, the real story isn't even really consumer, it's Muse for small businesses.

她写道，面向SMB的Muse可能是当前消费级智能体中最被低估的故事。

She wrote, Muse for SMBs might be the most underrated story in consumer agents right now.

他们以非常低的成本拥有近乎无限的消费者分发渠道，Meta旗下应用的日活跃用户基数达到36.4亿。

They have near-infinite consumer distribution at very low cost, with a base of 3.64 billion daily active users across Meta apps.

但更重要的是，超过5亿的SMB使用Meta进行营销。

But even more importantly, more than 500 million SMBs use Meta for marketing.

如果Meta为这些SMB群体提供智能体的可发现性，那将是一大推动，尤其是对本地商家而言，他们可以超精准地触达有紧急购买需求的用户。

If Meta gives agent discoverability to that SMB base, it could be a huge boost, especially for local businesses who can hyper-target users with urgent purchase needs.

Meta在第三季度财报电话会议以及此前的博客文章中表示，他们计划为SMB和Muse构建一个连接器生态系统。

Meta said in their Q3 earnings call and in blog posts prior that they plan to build an ecosystem of connectors for SMBs and Muse.

跨Meta账户的分析，再加上QuickBooks、Canva、Stripe等等。

Analytics across your Meta accounts, plus QuickBooks, Canva, Stripe, et cetera.

许多这类企业已经信任Meta，将他们的广告支出增长寄托在Meta上。

Many of these businesses are already trusting Meta with their growth in ad dollars.

我可以看到Muse几乎成为SMB的中枢神经系统，让Meta能够随着时间推移捕获他们越来越多的活动。

I could see Muse becoming almost a nervous system for SMBs that allows Meta to capture more and more of their activity over time.

有趣的是，正当关于个人智能体可行性的讨论沸沸扬扬之际，我们迎来了两个新的基准测试，它们在一定程度上能衡量这类个人智能体的价值。

Interestingly, even as these discussions about the viability of personal agents were swirling, we got two new benchmarks introduced that go some way towards measuring the value of these sort of personal agents.

第一个分析基准叫做GDPx，专注于具有经济价值的任务，比如芯片设计、金融和生物学。

The first analysis benchmark is called GDPx and is focused on economically valuable tasks like chip design, finance, and biology.

他们发现Opus 5.5是驱动智能体最有价值的模型，在他们的评分体系上得了12%，每项任务成本51美元。

They found that Opus 5.5 was the most valuable model to drive an agent, scoring 12% on their scale at $51 per task.

Gemini 4 Carbon排在第二位，得分10.9%，但每项任务只需16美元。

Gemini 4 Carbon was in second place at 10.9%, but only cost $16 per task.

Chubby在X上直接回应了这个效率问题。

Chubby on X jumped right at this question of efficiency.

他写道，虽然性能只差1.1%，但Gemini 4的价格大约是Opus 5.5的四分之一。

He wrote, While performing just 1.1% worse, Gemini 4 is about 4x cheaper than Opus 5.5.

所以从性价比来看，Gemini 4完胜。

So cost-performance-wise, Gemini 4 wins by a lot.

其中一个很明显的问题，实验室现在肯定在思考的，就是如何利用更高效的模型来降低个人智能体的成本，就像今年他们在编程智能体上一直重点做的那样。

And one of the obvious questions that clearly the labs are thinking about right now is how to use more efficient models to drive down the costs of personal agents, similar to what has been a focus with coding agents across this year.

等会儿我们会看到，现在也有各种各样的新训练技术的讨论，有些人觉得明年有可能会出现更强但更小的模型，让个人智能体变得划算得多。

As we'll see later, there's all sorts of buzz around new training techniques as well that lead some to think that it's possible that we'll see stronger, smaller models in the coming year that make personal agents a lot more cost-effective.

与此同时，OpenAI 发布了一个新基准，叫 Lifeval，目的是衡量个人智能体在日常生活中的价值。

OpenAI's new benchmark, meanwhile, called Lifeval, sets out to measure the value of personal agents in everyday life.

OpenAI 写道：对于目前每周使用 ChatGPT 的超过十亿用户，我们希望构建能帮助解决日常生活实际挑战的 AI，比如找托儿、获取政府服务、做预算、准备上学、调整食谱、买杂货、设计锻炼计划等等。

Writes OpenAI, For the more than 1 billion people who currently use ChatGPT every week, we want to build AI that can help with the practical challenges of everyday life, like finding childcare, accessing government services, creating a budget, preparing for school, modifying recipes, ordering groceries, designing workout plans, and more.

OpenAI 的评分显示，从 GPT-5.2 到 GPT-6 Astra，能力稳步提升，成功率从百分之四十九点三上升到百分之八十点六。

OpenAI's scoring shows a steady increase in capability from GPT-5.2 to GPT-6 Astra, rising from 49.3% success to 80.6% success.

更重要的是，OpenAI 现在开始以免费用户为对象，衡量这些日常任务并朝着它们进行爬山优化。

More importantly, OpenAI is now measuring and hill-climbing against these everyday tasks with free users in mind.

就像我们在编程领域看到的，一旦某个能力被衡量并成为强化学习的目标，它就会大幅提升，并且变得更加划算。

As we've seen with coding, once a capability is measured and becomes a target for reinforcement learning, it dramatically improves and becomes more cost-effective.

先忘了那些花哨的 AI 基准吧。

Forget all the flashy AI benchmarks.

对百分之九十九的 AI 用户来说，Lifeval 可以说是最重要的基准。

Lifeval is arguably the most important benchmark for 99% of AI users.

这才是真正重要的东西。

This is what actually matters.

感觉这种帖子就是在向公众发布个人智能体之前发出来的。

Feels like this is the kind of post you publish right before releasing a personal agent to the general public.

我关心的是，在 Muse 全面推出语音模式之前，ChatGPT 能不能先变得更擅长做个人智能体。

The question for me is whether ChatGPT gets better at being a personal agent before Muse gets widespread voice mode.

ChatGPT 的语音模式很棒，但 Muse 是更好的个人智能体。

ChatGPT's voice mode is awesome, but Muse is the better personal agent.

对了，我在工作中用 Dots，也试过在个人场景用，所以我知道它们的能力，但还是更喜欢 Muse。

And yes, I use Dots at work and tried them for personal use, so I know what they're capable of and still prefer Muse.

现在咱们来看看这个新闻，它乍一看好像就是今天 AI 圈子里最大的事。

Now let's actually move over to the story that with just a glance at the AI community seems to be the biggest one of the day.

你会发现，它可能对个人智能体相关的所有成本问题都有一些影响。

As you'll see, it potentially has some implications for all of these cost questions around personal agents.

长话短说，MIT 的研究人员发表了一项突破性研究，可能会改变模型训练的方式。

The TL;DR is that MIT researchers have published groundbreaking research that could change the way that model training works.

在博士生 Rukia Takahashi 的带领下，研究人员发现 GPT-6 Astra 能够读取并直接调整一个小模型的权重，从而提升它的性能。

Led by PhD student Rukia Takahashi, the researchers found that GPT-6 Astra was capable of reading and directly adjusting the weights of a small model to improve its performance.

实验用的是 Qwen3.5-0.8B，目标是改善推理能力。

The experiment used Qwen3.5-0.8B and targeted improvements in reasoning.

Astra 只改了百分之零点零一六的权重，就让 Qwen 在数学基准上的得分提升了六倍，指令遵循能力提升了三倍。

Astra changed only 0.016% of the weights, and yet was able to improve Qwen's scores on math benchmarks by 6x and instruction following by 3x.

在他们的摘要中，研究人员写道：我们发现，前沿大语言模型在原生读取和写入神经网络权重方面出奇地擅长，远超我们预期的人类或传统算法。

In their abstract, the researchers wrote, we find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we'd expect from humans or conventional algorithms.

这对优化和可解释性来说，可能都是件大事。

This might be a big deal for both optimization and interpretability.

仅仅通过看权重，GPT-6 Astra 就能判断出一个网络是训练来做什么的。

Just by looking at the weights, GPT-6 Astra can tell what a network was trained to do.

而且不需要任何环境反馈，它就能毫无困难地从零写出小型玩游戏网络，这些网络的表现甚至超过了经过 PPO 训练一亿步的网络。

And without any environment feedback, it has no trouble writing, from scratch, small game-playing networks that outplay networks trained with PPO for 100 million steps.

顺便说一句，PPO，也就是近端策略优化，是各大实验室都在用的标准强化学习算法。

PPO, or Proximal Policy Optimization, by the way, is a standard reinforcement learning algorithm in use by the major labs.

这篇研究值得完整阅读，或者让 Claude 或 ChatGPT 读一下然后跟它讨论，里面包含多个令人惊讶的结果。

The research is worth reading in full, or having Claude or ChatGPT read and talking to it about, and includes multiple surprising results.

但说到修改 Qwen 的权重，这个过程极其直接。

But when it comes to altering Qwen's weights, the process was extremely straightforward.

我们只是让 Astra 把 Qwen3.5-0.8B 变得更擅长数学。

We simply asked Astra to make Qwen3.5-0.8B better at math.

Astra 发现 Qwen 陷入推理循环——这是 Qwen 自己的模型卡里警告过的已知问题——在练习题上用五十次前向传播的预算，就找到了原因并将其编辑掉了。

Astra saw Qwen get stuck in reasoning loops, which is a known issue that Qwen's own model card warns about, With a budget of 50 forward passes on practice problems, it located the cause and edited it out.

这个过程没有用到梯度或训练。

This used no gradients or training.

通过只编辑百分之零点零一六的权重来修复这些循环，还顺带提升了数学以外的其他能力。

Fixing these loops by editing just 0.016% of weights boosted other capabilities beyond math too.

大家对这个结果兴奋得不得了。

People were incredibly excited about this result.

Yuchen Jin 写道，这也太酷了吧！

Yuchen Jin wrote, this is cool as heck!

想象一下，用一个大 LLM 直接去编辑一个小模型的权重，然后大幅提升它的能力。

Imagine using a large LLM to directly edit a smaller model's weights and dramatically improve its capabilities.

这就像是给神经网络做基因编辑。

It's like gene editing for neural networks.

比蒸馏还要酷。

Even cooler than distillation.

开发者 Nick Dobos 写道，Astra 能从零开始手写出新的权重，而且效果还超过了其他 AI 训练方法？

Developer Nick Dobos wrote, Astra can handwrite new weights from scratch that outperform other AI training techniques?

什么？

What?

他们把重点给埋没了！

They buried the lead!

通过修复循环 bug 来提升数学能力确实很酷，但调试和修复权重，跟从零生成全新权重比起来，还是差得远。

Improving math via a looping bug fix is very cool, but debugging and fixing weights is quite limited versus generating brand new weights.

如果这真的行得通，那以后还需要训练大型 AI 模型吗？

If this works, is training large AI models even needed anymore?

这算真正的 RSI 吗？

Is this true RSI?

而对于像 Ethan Mollick 教授这样的人，他说，这看起来确实很酷，但咱们还是等一些可复现的结果吧——好吧，没过多久，Google 的资深研究员 Alex Krentzel 就复现了这个结果。

And for those like Professor Ethan Mollick who said, this seems really cool, but let's wait for some replicated results, well, it didn't take long for Alex Krentzel, a senior researcher at Google, to replicate the results.

Alex 指出，这是一种新出现的、跨越式的能力跃升，而且只出现在最新一批模型里。

Alex noted, this is a new step-change emerging capability that's appeared in just the latest set of models.

不过他也接着说，有一些非常重要的注意事项，得先搞清楚才能理解这些结果。

Now, he did continue, some very important caveats to understand these results.

注意，这里被编辑的模型都非常小。

Note that the models being edited here are very small.

他们文章里和我实验里那个最出彩的模型，只有两万零四百八十六个参数。

The breakout model in their post and my experiment is 20,486 parameters.

而前沿模型，参数量超过一万亿。

Frontier models are more than a trillion parameters.

我的理解是这样的：2010 年代有一大批可解释性研究，比如，看这个模型，这张图里有狗的时候这个神经元就亮了，所以这就是“狗神经元”。

The way that I'm thinking about it is this: there's a whole body of work of interpretability research from the 2010s, à la, look at model, this neuron lights up when image contains dog, so this is the dog neuron.

这件事跟那个很像，只不过规模已经超出了人类手工能做到的程度。

This seems akin to that, but scaled beyond human capability.

以前我们对小模型是手工做这些的，然后我们造了工具，帮助在稍微大一点的模型上做。

We've done this by hand for smaller models, then we built tools to help this for slightly bigger models.

现在我们看到的是，推理能力让模型能够对更大、但仍然相对较小的模型手工完成这件事。

What we're seeing is the reasoning capabilities are enabling the models to do this by hand for much larger, but still relatively small models.

Krentzel 还说，这项研究显而易见的下一步，是测试运行时自我改进。他认为这对当前的模型来说还不可行，但也许再过几代就能做到。

Krentzel added that the obvious next step in the research is to test self-improvement at runtime, which he believes won't be feasible for current models but could work within a few generations.

你可以想象，每当出现这种大张旗鼓、夸张的说法，总会有一场相当激烈的争论，讨论这些结果是不是被过度炒作了。

As you might expect, whenever you have big, bombastic claims like this, there's going to be a pretty heated debate about whether the results were overhyped.

来自 Voyage AI 的 Chandan Singh 评论说，对于那些质疑这篇文章是不是过度炒作的人，我觉得他们其实还低估了这个方法。

Chandan Singh from Voyage AI commented, for anyone questioning whether this post is overhyped, I think they are actually underselling the approach.

我们做了类似的实验，发现 Claude 模型能够从零写出完整的、可以正常工作的神经网络权重。

We did similar experiments and found that Claude models can write entire, functioning neural net weights from scratch.

不过另一边也有人，比如微软研究院的 Weiya Zhu，他说，这些发现确实挺有意思，但我觉得标题有点过度炒作了。

On the other side were people, however, like Weiya Zhu from Microsoft Research who said, the findings are pretty interesting indeed, but I think the headline is kinda overhyped.

数学上六倍的提升，来自修复了原模型里的一个循环 bug。

The 6x improvement on math came from fixing a loop bug in the original model.

而不是来自通用数学能力的提升。

Not from improving general math capability.

而这正是主要的批评点——这与其说是一种新的模型训练技术，不如说更像是给现有模型修了个 bug。

And this was the main critique— that this is less about a new model training technique, and more akin to a bug fix in an existing model.

然而对另一些人来说，这却是明确的证据，说明递归自我改进不仅是未来的可能性，现在还有了清晰的机制——这个机制已经被理解，并且已经在一篇学术论文中被演示出来了。

And yet for others, this was clear evidence that recursive self-improvement is not only a possibility in the future, but now has a clear mechanism that's understood and has been demonstrated in an academic paper.

我认为你应该强烈预期这类事情会在未来一年内发生，而且它将成为 RSI 的驱动因素之一。

I think you should strongly expect something like this to happen in the next year and for it to be one of the driving factors of RSI.

而且可能是相当陡峭的 RSI。可解释性具有极强的双重用途——擅长大规模优化的模型，也很擅长优化自己。

And potentially quite steep RSI, interpretability is incredibly dual-use— models that are good at optimization writ large are good at optimizing themselves.

现在，RSI 也是围绕 Anthropic 周一早上发布的 S-1 更新文件讨论的主要焦点。

Now, RSI was also the main point of discussion around an update of Anthropic's S-1 that they released on Monday morning.

AI 新闻聚合者 Andrew Curran 写道，Anthropic 今早在 S-1/A 文件中新增的措辞说，内部估计表明他们可能即将达到 RSI 的危险阈值——即归因于 AI 的进展将急剧加速，一年取得的进展相当于 2018 到 2024 年基线两年的进展。

AI news aggregator Andrew Curran wrote, New language added on Anthropic's S-1/A this morning says that internal estimates indicate they may imminently meet the RSI danger threshold of dramatically accelerated progress attributable to AI, as much progress in 1 year as 2 years of the 2018-2024 baseline.

他们还补充说，正在考虑更新其负责任扩展政策，因为他们“不预期能够在没有有意义的人类监督或理解的情况下，安全地使用 Claude 完全自主地设计和构建连续几代前沿模型。”

They added that they are considering updates to their responsible scaling policy because they, quote, do not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding.

基本上，这些新变化的目标是让人类继续掌控前沿模型的开发，而不是把研究完全交给模型。

Basically, the new changes will be aimed at keeping humans in control of frontier model development rather than handing research over to the model entirely.

当然，金融专业人士确实在关注 RSI，把它视为这次 IPO 最重要的因素之一。

Certainly it is the case that financial professionals are taking note of RSI as one of the most important factors for the IPO.

《华尔街日报》发表了一张来自 Anthropic 的图表，显示 AI 现在主导了 26% 的研究。

The Wall Street Journal published a chart from Anthropic showing that AI is now leading 26% of research.

他们的标题写道：AI 领域最聪明的人认为，这是世界上最重要的一张图表。

Their headline read, the smartest people in AI believe this is the world's most important chart.

AI 正在开始构建更好的 AI。

AI is starting to build better AI.

这种自我改进正在加速研发，也让所有人都慌了。

This self-improvement is speeding up research and development and freaking everyone out.

这个标题似乎是在引用 Anthropic 联合创始人 Jack Clark 的话，他说 RSI“可能是世界上最重要的技术趋势。”

The headline seemed to reference Anthropic co-founder Jack Clark, who said that RSI, quote, might be the most important technical trend in the world.

顺便说一句，Anthropic 最早是在九月份基于八月份的数据发布这张图表的。

Anthropic first published this chart, by the way, back in September based on August numbers.

而且上周末，前 AI 事务负责人 David Sacks 还转发了它。

And it was boosted by former AI czar David Sacks over the weekend.

不过，随着 AI 叙事开始变味，或者至少像我们一直探讨的那样出现了更多疑问，值得注意的是，华尔街投资者正在关注 RSI 可能导致模型能力加速提升这一可能性。

Still, with the AI narrative beginning to sour, or at least having more questions as we've been exploring, it's notable that Wall Street investors are paying attention to the possibility that RSI leads to faster advances in model capability.

正如 Sacks 评论的那样，那还是八月份的事，所以谁知道到了十月份他们现在是什么情况呢。

As Sacks commented, that was back in August, so who knows what they're at now, being in October.

现在，我觉得那场讨论中有趣的一点是这样一个预设——它如此根深蒂固，以至于没人觉得需要说出来——那就是，提高采用率的关键是更好的模型。

Now, one thing I find interesting about that discussion is the presumption, so deeply embedded that no one feels the need to say it, that the key to greater adoption is improved models.

我们拥有的证据是，过去真正的阶跃式变化确实解锁了新能力，并由此带来了采用量的大幅增长。

The evidence that we have for that is that true step changes have unlocked new capabilities that have led to massively more adoption in the past.

其中最显著、最重要的，当然是 Opus 4-5 的跃升，它引领了我们此后一直身处的 full-agent decoding 时代。

The most notable and important of these, of course, was the Opus 4-5 jump that led to the full-agent decoding era that we've been living in ever since.

不过，在没有某种全新能力上线的情况下，我不认为证据表明提高采用率的关键仅仅是模型能力。

In the absence, though, of some entirely new capability coming online, I don't think the evidence suggests that the key to improved adoption is solely model capability.

我认为归根结底，这些商业模式是否行得通，或多或少取决于企业消耗多少智能。

I think ultimately, when it comes to whether these business models work or not, it's more or less a question of how much intelligence enterprises consume.

目前，他们消耗的只是他们本可以消耗的极小一部分，而这并不是因为模型能力不够高。

They are, at the moment, consuming a vanishingly small portion of what they could, and it's not because the model capabilities aren't high enough.

而是因为惯性是真实存在的，企业系统笨重迟缓，而采用本身就很困难。

It's because inertia is real, corporate systems are lumbering, and adoption is hard.

现在，我觉得基于这一点，你可以写出看空理由，也可以写出看多理由。

Now, I think you could write a bear case or a bull case based on that.

看空的理由是：即使拥有所有这些能力，我们也没有在使用它。

The bear case is that even with all this power, we're not using it.

看多的理由是：不，我们只是还没开始用而已。

The bull case is, No, we're just not using it yet.

无论如何，投资者想怎么对自己解释 RSI 及其对采用的影响都行。

Regardless, whatever investors want to tell themselves about RSI and its impact on adoption is fine.

我只是不确定，模型改进是否像我们直觉上以为的那样，能一条直线地通向收入增长。

I'm just not sure model improvement has quite such a straight line to increased revenue as we might intuitively think.

我想说的是，至少其中一些投资者开始问：当我们真正拥有 RSI 时会发生什么，而不只是我们离它还有多近？

What I will say is that at least some of those investors are starting to ask, what happens when we have RSI, not just how close to it are we?

感觉我们仍然卡在原地，大体上还是在讨论RSI到底来了没有，而不是讨论它到来时意味着什么。

It feels like we're still stuck, broadly speaking, is RSI here yet or isn't it discussion rather than, What it means when it arrives.

目前，唯一对RSI之后的世界有所愿景的群体，是那些恐惧它、认为它应该被定为非法的AI安全主义者。

Right now, the only group offering any sort of vision for the world of RSI is the AI safetyists who are terrified of it and think it should be illegal.

我们可能不认同他们，但至少他们确实在阐述，他们认为如果RSI到来，世界会变成什么样。

We may not agree, but at least they're actually articulating what they think the world is going to look like after that, if and after RSI comes.

让我触动的是，就像之前的AGI一样，我们可能被这个三个字母的缩写词搞得过于分心，花了更多注意力去定义它、说明它离我们有多近，而不是去理解它到来时意味着什么。

It strikes me that, like the AGI term before it, we may be getting so distracted with this 3-letter acronym that we're spending more attention on trying to define it and articulating how close it is to being here than understanding what it will mean when it arrives.

不管怎样，AI领域依然是一个令人着迷的时代，但今天的AI每日简报就到这里。

In any case, it continues to be a fascinating time in AI, but that is gonna do it for today's AI Daily Brief.

照例感谢大家的收听或收看，下次见，祝平安。

Appreciate you listening or watching as always, and until next time, peace.

已剔除 4 处广告（点击展开查看）

M10:00广告 · 已剔除

Before we get into today's episode, I wanted to share that the second episode of How We Got Here has arrived.

The episode is called How We Made the Wrong Things Cheap and looks at how for 25 years the price of stuff— consumer goods, electronics, etc.— went down while the cost of housing, healthcare, and all that really important stuff went the other direction.

I think in many ways this dichotomy is at the heart of the economic populism that is increasingly dominating American politics.

For those of you who have already listened and subscribed to How We Got Here, a big thank you.

And for those who haven't yet, this is my invitation.

If you like what you get on the AI Daily Brief, I think you'll like this show.

So go check it out, How We Got Here with NLW, on any podcast platform, or go to howwegothere.com.

M11:17广告 · 已剔除

First of all, thank you to today's sponsors, KPMG, Robots and Pencils, ZData, and Blitzy.

To get an ad-free version of the show, go to patreon.com/AIDailyBrief, or you can subscribe on Apple Podcasts.

And to learn more about sponsoring the show, send us a note at [email protected].

M11:34广告 · 已剔除

As you just heard, the 2nd episode of How We Got Here is now out.

It's called How We Made the Wrong Things Cheap.

And for any of you who are interested in the economics of the modern American economy, whereby some measures we're doing better than ever, but everyone feels more behind, If you wanna know how that came to be, that episode is for you.

You can find it at howwegothere.com or by searching How We Got Here with NLW on any podcast platform.

M111:38广告 · 已剔除

One big change around AI is we've shifted our thinking from how we rank our pages to how do we become the source that AI trusts enough to answer with.

At KPMG, they're seeing this firsthand.

AI-generated results now surface answers directly, often without a single click.

That's why they are increasingly focused on generative engine optimization, or GEO, structuring content so AI systems can retrieve it, understand it, and cite it as a trusted authority.

This is not just an SEO evolution, but a visibility mandate.

And indeed, the GEO mandate from KPMG is simple.

If AI is shaping decisions, your expertise needs to show up inside the answer.

Read all about it at kpmg.com/us/geo.

Again, that is kpmg.com/us/geo.

The best teams don't have a single star carrying everyone else.

They know their own strengths and each other's weaknesses and play to both.

That's the team Robots and Pencils has built on purpose.

Nobody there is grinding through busywork to pad a headcount number.

People come for the hard problems, and they stay because everyone around them is leveling up at the same time.

In a market full of companies that are just trying to hire fast, that's worth a look.

Check out robotsandpencils.com/careers.

This episode is brought to you by CData.

Companies rolling out AI keep hitting the same wall.

Token spend is climbing, security is holding agents back, and no one knows how AI is truly being used.

The root cause: AI is cut off from the systems and context your business runs on.

That's why CData built Connect AI Gateway, one control point between your AI and your systems.

Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more, each agent gets exactly the data and tools it needs.

It governs down to the record and lets you finally see your org's complete AI usage in one place.

Accurate answers, a fraction of the token cost, and a gateway that learns your business with every interaction.

See how at cdata.ai.

Blitzy deeply understands your codebase before it writes code.

Here's the first place that pays off: security in the age of AI.

Vulnerabilities don't live in isolation.

They live buried inside millions of lines of interconnected code, where patching one thing quietly breaks 3 others.

That's why surface-level scans fail.

Blitzy starts from its knowledge graph of your entire application, identifies and surfaces CVEs across the full estate, proactively recommends patches, and can execute the PR.

Each fix is grounded in how your systems connect and validate so nothing new breaks, and the knowledge graph dynamically updates, keeping you ahead of an ever-accelerating threat landscape.

One Blitzy customer resolved 21 active CVEs across 6 core microservices in 4 days.

Zero compile errors, every validation scan clean, months of planned work fixed in less than a week.

Security remediation grounded in real architectural context at the speed of compute.

Harden your codebase at blitzy.com.

That's B-L-I-T-Z-Y dot com.