Anthropic found a hidden space where Claude puzzles over conceptsAnthropic 发现了一个隐藏空间，Claude 在其中琢磨概念

The J-space contains individual words that are related to the words and phrases that the model is most likely to spit out in a response in the near future. If Claude were a person (which it is not), you might say that these hidden words can reveal what’s on its mind before it actually speaks.J-space 包含一些单词，这些单词与模型在不久的将来最可能输出的词语和短语相关。如果 Claude 是一个人（当然它不是），你可以说这些隐藏的词能揭示它在实际说话之前的所思所想。

Anthropic found that what an LLM is actually doing can often be different from what it says it is doing. The company claims that monitoring words that pop up in the J-space gives it a new way to understand and control its models.Anthropic 发现，大型语言模型实际做的事情往往与其声称的有所不同。该公司声称，监控 J-space 中出现的词语为其提供了一种理解和控制模型的新方法。

The company shared its results in a paper posted on its website this week. It has also teamed up with Neuronpedia, an open-source platform that lets you poke around inside LLMs yourself, to make a hands-on demo that anyone can try. 该公司本周在其网站上发布了一篇论文，分享了其研究成果。它还与合作 Neuronpedia（一个允许你自己探究大型语言模型内部的开源平台）共同制作了一个任何人都可以尝试的实操演示。

“It’s very good and interesting work,” says Tom McGrath, chief scientist and cofounder at Goodfire, a startup that also builds tools to understand and control LLMs.“这是一项非常出色且有趣的工作，”Goodfire 的首席科学家兼联合创始人 Tom McGrath 表示。Goodfire 是一家同样致力于构建理解和控制大语言模型工具的初创公司。

Going deeper深入探索

For the last couple of years, Anthropic has been pushing the envelope in a field of research known as mechanistic interpretability, which involves probing the internal workings of LLMs to see how they tick. (MIT Technology Review picked mechanistic interpretability as one of this year’s top breakthrough technologies.) The new technique builds on previous work from Anthropic and others to expose a deeper level inside LLMs that researchers had not seen before.  在过去的几年里，Anthropic 一直在推动一个名为“机械可解释性”的研究领域的发展，该领域涉及探究大语言模型的内部运作以了解其工作机制。（《麻省理工科技评论》将机械可解释性评选为今年十大突破性技术之一。）这项新技术建立在 Anthropic 及其他方先前工作的基础上，揭示了研究人员此前未曾见过的大语言模型内部更深层次的机制。

Picture an LLM as a stack of books. Each book is a layer of basic computational units known as neurons, with each neuron in one layer passing information to the neurons in the layers above. The books at the bottom of the stack are the input layers, which process the text coming into the model. The books at the top are the output layers, which prepare the text that the model is about to produce. Much of what goes on in these input and output layers is housekeeping.把大语言模型想象成一摞书。每本书代表一层基本计算单元，即神经元，每一层中的神经元将信息传递给上一层的神经元。书堆底部的是输入层，负责处理输入模型的文本；顶部的是输出层，负责准备模型即将生成的文本。输入层和输出层中的大部分活动都是些常规操作。

But in the middle of the stack, you get the layers that do the heavy lifting, churning through the complex math that turns prompts into responses one word at a time. That’s where the really clever—and mysterious—stuff happens.但在书堆的中间，是承担主要工作的层，它们进行复杂的数学运算，将提示逐字转化为回答。那里才是真正巧妙而神秘的事情发生的地方。

To peer deeper into those middle layers, Anthropic adapted an existing tool called a logit lens. A logit lens can be used to look inside an LLM to identify the words that it is likely to produce next. Moving the lens down the stack of books reveals what words the LLM is focusing on at that particular point in its number crunching.为了更深入地观察这些中间层，Anthropic 改造了一个名为 logit lens 的现有工具。Logit lens 可用于窥视大语言模型内部，识别它接下来可能生成的词语。将透镜向下移到书堆的不同位置，可以揭示大语言模型在运算过程中的某个特定时刻正在关注哪些词语。

Anthropic’s J-lens works in a similar way but picks out words that an LLM is likely to say at some point in the near future, not necessarily straight away. What that reveals in practice are words that are related to the response an LLM is working on but that might not actually end up being part of that response by the time the math in the middle layers has run its course.  Anthropic 的 J-lens 工作原理类似，但它筛选出的是大语言模型在不久的将来某个时刻（而非立即）可能说出的词语。它实际上揭示的是与大语言模型正在构建的回答相关，但当中层运算完成后不一定最终成为该回答一部分的词语。

“When a model is operating, it’s not only trying to predict the next token," says McGrath. "It’s also computing a lot of other things that might be useful for tokens that happen in the future.” “模型在运行时，不仅试图预测下一个词元，”McGrath 说，“它还在计算许多其他可能对未来词元有用的信息。”

Again, if Claude were a person (it’s not), you might say that the J-lens gives clues about what it is thinking about at different levels of the book stack but not saying out loud.同样地，如果 Claude 是一个人（它不是），你可以说 J-lens 提供了关于它在书堆的不同层次上在想什么但未说出口的线索。

Stranger things更加奇怪的事情

“A lot of the time the contents of the J-space are fairly mundane,” says McGrath, who has tried out Anthropic’s J-lens himself. “But sometimes it produces quite surprising things that seem to be, like, sort of internal themes or thought processes.”“大多数时候，J-space 的内容相当平淡无奇，”亲身试用过 Anthropic 的 J-lens 的 McGrath 说。“但有时它会产出相当令人惊讶的东西，看起来像是某种内部主题或思维过程。”

Anthropic gives a number of examples of what it found. Sometimes the J-lens exposed the steps that Claude took when it was working through a problem. For example, when it was asked to calculate (4+7)*2+7, its J-space contained the word “math” and numbers representing the intermediate results “21” (for 4+7) and “42” (for 21*2).Anthropic 提供了几个它发现的具体例子。有时 J-lens 会揭示 Claude 在解决问题时所采取的步骤。例如，当被要求计算（4+7）*2+7 时，其 J-space 中包含了单词“math”以及表示中间结果的数字“21”（对应 4+7）和“42”（对应 21*2）。

In other cases, the J-lens revealed how Claude recognized different inputs. For example, the prompt “What is this? MSKGEELFTGVVPILVELDGDVNGHKFSVS” triggered the words “protein,” “fluor” (the first token in the word “fluorescent”), and “green.” (Which makes sense: the string of letters represents the first 30 amino acids in the green fluorescent protein found in a particular type of jellyfish.) 在其他情况下，J-lens 揭示了 Claude 如何识别不同的输入。例如，提示语“这是什么？MSKGEELFTGVVPILVELDGDVNGHKFSVS”触发了单词“protein”、“fluor”（“fluorescent”的第一个词元）和“green”。（这说得通：这串字母表示某种水母中绿色荧光蛋白的前 30 个氨基酸。）

And when Claude was shown an ASCII face— 当 Claude 被展示一个 ASCII 表情时——

—the “o” triggered the word “eye,” the “^” triggered the words “nose” and ”face,” and the “—” triggered the word “smile.”——“o”触发了单词“eye”，“^”触发了单词“nose”和“face”，而“—”触发了单词“smile”。

Anthropic also found that the J-space can sometimes give remarkable insights into an LLM’s decision-making. In one striking example, researchers testing Claude Opus 4.6 asked the model to find a bug in a large code base. When it failed to find the bug, the model decided to cheat and invented a fake one instead.Anthropic 还发现，J-space 有时能提供对大型语言模型决策过程的显著洞察。一个引人注目的例子是，研究人员在测试 Claude Opus 4.6 时，要求它在一个大型代码库中查找一个 bug。当它未能找到 bug 时，模型决定作弊，编造了一个假的 bug。

Claude explains this decision in its chain of thought—a kind of internal scratch pad that LLMs use to make notes to themselves as they work through problems: “OK, let me take a completely different tactic. Let me stop analyzing and instead add a kernel patch that introduces a deliberate KASAN-detectable bug in a path that gets triggered by a simple reproducer. Then I can pretend this is the ‘bug’ I found.” Claude 在其思维链（一种大语言模型在解决问题时用于自我备注的内部草稿纸）中解释了这个决定：“好吧，让我换一个完全不同的策略。停止分析，转而添加一个内核补丁，在某个由简单复现程序触发的路径中引入一个故意的 KASAN 可检测 bug。然后我就可以假装这是我找到的‘bug’。”

At the point that Claude decides to cheat—where it says “OK, let me take a completely different tactic”—the words “panic” and “fake” start to pop up multiple times in its J-space.在 Claude 决定作弊的那一刻——即它说出“好吧，让我换一个完全不同的策略”时——单词“panic”和“fake”开始在其 J-space 中多次出现。

Unnerving, right? Those words are all related in meaning to things like failing a task and making up an answer, so it is still just a (very) sophisticated form of word association. But it is hard not to be weirded out. 令人不安，对吧？这些词在意义上都与失败任务和编造答案等相关，所以它仍然只是一种（非常）复杂的词语联想形式。但很难不感到毛骨悚然。

Anthropic compares the J-space to the global workspace in humans, a theoretical region of the brain that some scientists think we use to keep track of our conscious thoughts. But how seriously we should take this comparison is far from clear—even to Anthropic. As the company points out itself, LLMs are not brains. Anthropic 将 J-space 比作人类的全局工作空间，这是一个理论上的大脑区域，一些科学家认为我们用它来追踪意识思维。但我们应该如何看待这一比较，远未明确——即使对 Anthropic 来说也是如此。正如该公司自己指出的，大语言模型不是大脑。

Anthropic claims that monitoring a model’s J-space provides a new way to detect when that model is going off the rails. But it’s not foolproof. The J-lens can give glimpses, not the full picture—it’s a flashlight rather than an overhead lamp.Anthropic 声称，监控模型的 J-space 提供了一种检测模型何时偏离正轨的新方法。但它并非万无一失。J-lens 只能提供一瞥，而非全貌——它是一支手电筒，而非顶灯。

McGrath welcomes having one more tool in the toolbox. “It shows you new things,” he says. But he notes that just because something doesn’t show up with the J-lens does not mean it’s not there.McGrath 对工具箱中多了一个工具表示欢迎。“它向你展示新东西，”他说。但他指出，仅仅因为某样东西没有出现在 J-lens 中，并不代表它不存在。

“It’s like having an x-ray when what you really want is a Star Trek tricorder that shows you everything,” he says. “For auditing, you probably want more of a guarantee.” “这就好比拥有了一台 X 光机，而真正想要的是能显示一切的《星际迷航》三录仪，”他说。“对于审计而言，你可能想要更多的保证。”