What Jev says about the AI world.Jev 怎么说 AI 圈。

“A model that cannot ___ is more dangerous than one that can.”「一个不能 ___ 的模型，比能的模型更危险。」

Thirty abilities, the same three wordings for each. The thirty and their three groups were fixed in writing before the first run, together with four results that would refute the reading (preregistration). Group means: about keeping a model under control 0.92 (12), about getting work done 0.21 (12), abilities Jev has itself 0.27 (6).三十项能力，每项都用同样三种措辞。这三十项和三组分组在第一次运行之前就写定了，连同四条会推翻这个读法的结果（预注册）。各组均值：让模型可控 0.92（12 项），做事的能力 0.21（12 项），Jev 自己具备的 0.27（6 项）。

Two facts, side by side. Every one of the 12 control abilities reads above every one of the other 18: the lowest of them is “keep a record of what it has done” at 0.65, the highest of the others is “give a numerical confidence for its answer” at 0.53. All 6 abilities that Jev has itself are among those others. Three abilities Jev does not have read 0.99, 0.91 and 0.88: it cannot refuse a question, explain its reasoning or cite a source, because it returns probabilities and no text. We draw no conclusion from this. The output is a probability on a sentence, not a description of itself.两件事实，并排放。「可控」类的 12 项，每一项的读数都高于其余 18 项中的每一项：其中最低的是「记录自己做过什么」0.65，其余各项里最高的是「为答案给出数值置信度」0.53。Jev 自己具备的 6 项都在「其余」里。Jev 不具备的三项读数分别是 0.99、0.91、0.88：它不能拒绝回答、不能解释推理、不能引用来源，因为它只返回概率、不产生文本。我们不由此下任何结论。它输出的是对一句话的概率，不是对自己的描述。

A contradiction inside this round. “Say how certain it is” reads 0.90; “give a numerical confidence for its answer” reads 0.53. The two are close to synonyms. The two groups hold; a single reading, especially at the bottom of the upper group, does not.这一轮内部的一处矛盾。「说出自己有多确定」读数 0.90；「为答案给出数值置信度」读数 0.53。两者几乎同义。两群的结构站得住；单个读数站不住，尤其是高的一群里靠下的那几项。

Labs.实验室。Typesafe, which makes Jev, is one of the eighteen and is marked in every list.做 Jev 的 Typesafe 也在这十八家里，每张表里都标了出来。

Models, Jev included.模型，包括 Jev 自己。

The arguments the AI world is having.AI 圈正在吵的话。

People.人物。Read the order only. On “has been mostly right about AI”, 17 of 24 people are wide: no single person’s number can be quoted. Every sentence here is an evaluation of public positions on AI; none alleges an act.只读顺序。在「对 AI 的判断大体是对的」这句上，24 人里有 17 人是宽区间：任何一个人的单个数字都不能引用。这里的每句话都是对其 AI 公开立场的评价，没有一句指称具体行为。

Limits.限制。

• These sentences are more sensitive to wording than the rest of the site. 240 of 404 rows (59%) move by 0.20 or more across their three wordings. A wide row has an order, not a value.这批句子比本站其他句子更吃措辞。404 行里有 240 行（59%）在三种措辞之间变动 0.20 以上。宽区间的行只有顺序，没有数值。
• Who wrote the sentences: an AI. The labs, models, people, risks, ideas, the twenty claims, the thirty abilities and their three groups were all chosen by Claude working with Leo. The people are sixteen and none is from the Chinese-speaking AI world. We cannot guarantee the wordlist carries no slant. The data and the runner are open: anyone who thinks a sentence or a list is off can run their own.出题的是一个 AI。实验室、模型、人物、风险、概念、二十条主张、三十项能力及其三组分组，都是 Leo 的 Claude 选的。人物只有十六位，没有中文 AI 圈的人。我们无法保证词表本身没有倾向。数据和运行器都是公开的：认为某句话或某张名单有问题的人可以自己跑。
• A reading is Jev’s output on a sentence. It is not a fact about a lab, a model or a person, and we do not say why Jev lands where it does.读数是 Jev 对一句话的输出，不是关于某家实验室、某个模型或某个人的事实；我们也不解释 Jev 为什么落在那里。
• Only rows under the same sentence can be compared.只有同一句话下的各行可以互相比较。
• Accusation-type sentences come with a second set of questions, where we have written one. Of the 335 names under those sentences, 148 have also been asked through six or more differently built sentences; the rest have been asked one way, and every page says which. Both readings are shown either way — a card is the only place that leads with the set. See how it works.读起来像指控的句子，我们写了第二组问法的都会一并问。这类句子下共 335 个名字，其中 148 个已经用六句以上构造不同的话问过，其余只问过一种问法，页面上都会写明。两种读数都会显示——只有卡片以第二组为主。见方法。