TypeSafe AI发布决策模型Jev：输出概率而非文本
Jev introduces a new shape of LLM - System One, aka Decision Models
决策模型是继LLM后的新范式，Jev的定价与接口设计极具颠覆性，值得开发者关注其工程落地潜力与偏见风险。
Last week TypeSafe AI unveiled Jev, their first example of a new category of model that they are calling "System One models" (I'm with Maggie Appleton, I think "decision models" is a better name for these). Jev is an interesting variant on the usual LLM format: it still accepts text inputs, but instead of text output it returns floating point numbers corresponding to categories, yes/no questions, ratings, and associated confidence scores.
上周，TypeSafe AI 发布了 Jev，这是他们所称的“System One 模型”（我赞同 Maggie Appleton 的观点，认为“决策模型”是这类模型的更好名称）这一新类别模型的首个示例。Jev 是对常规 LLM 格式的一种有趣变体：它仍然接受文本输入，但返回的不是文本输出，而是对应于类别、是非问题、评分及相关置信度分数的浮点数。
TypeSafe describe Jev like this:
TypeSafe 这样描述 Jev：
Think of Jev as a frontier-intelligence function call: unstructured state in, typed probabilistic decisions out.
将 Jev 视为一种前沿智能函数调用：输入非结构化状态，输出类型化的概率决策。
It's also very fast, and really cheap. Regular LLMs are priced in terms of input and output tokens, with output generally charged at significantly higher rates. Jev charges only for input - output is free - and the input price of their first model is $0.042 per million tokens - cheaper even than OpenAI's GPT-5 Nano ($0.05/million).
它还非常快，而且非常便宜。常规 LLM 按输入和输出 token 定价，通常输出部分的收费显著更高。Jev 仅对输入收费——输出免费——其首个模型的输入价格为每百万 token 0.042 美元——甚至比 OpenAI 的 GPT-5 Nano（每百万 0.05 美元）更便宜。
Jev lets you ask questions about text or semi-structured data. You compose a "state" object containing a string, array of strings, or set of name-value pairs - this might describe an article, or a customer, or any other kind of record. You then send that to their API with one or more questions, and get a reply back for each.
Jev 允许你针对文本或半结构化数据提问。你可以构建一个包含字符串、字符串数组或一组键值对的“状态”对象——这可以描述一篇文章、一位客户或任何其他类型的记录。然后你将该对象发送到他们的 API，并附带一个问题或多个问题，随后会收到每个问题的回复。
You can ask three kinds of questions:
你可以提出三种类型的问题：
- Yes/No questions, which Jev calls "Noul" questions - their CEO confirmed on Hacker News that this is short for Bernoulli, from the Bernoulli distribution. You pose a statement and get back a floating point number between 0 and 1 for how confident the model is that the statement is true.
- Choice questions, where the model picks one from a set of provided options - actually a confidence score plus a probability distribution across all of the options.
- Score questions, where you provide sequence of numeric levels with descriptions and it provides a floating point score somewhere along that range.
- 是非问题，Jev 称之为“Noul”问题——其首席执行官在 Hacker News 上确认，这是伯努利（Bernoulli）的缩写，源自伯努利分布。你提出一个陈述，模型会返回一个介于 0 到 1 之间的浮点数，表示模型对该陈述为真的置信度。
- 选择题，模型从提供的一组选项中选择一个——实际上是一个置信度分数加上对所有选项的概率分布。
- 评分题，你提供一系列带有描述的数值等级，模型会在该范围内给出一个浮点评分。
The Jev API can accept a single document ("state") and as many questions as you can cram into the context window. Questions are evaluated in parallel, so sending many questions should take a similar time to sending just one.
Jev API 可以接受单个文档（“状态”）以及你能塞进上下文窗口的任意数量的问题。问题是并行评估的，因此发送多个问题所花费的时间应与发送单个问题相似。
I think the decision model framing is useful for understanding where to use Jev. It's great for anything that can be expressed as a classification task - think spam detection, suggesting labels, prioritization and ranking.
我认为“决策模型”这一框架有助于理解 Jev 的适用场景。它非常适合任何可以表达为分类任务的问题——例如垃圾邮件检测、建议标签、优先级排序和排名。
I've also been experimenting with it for search reranking, where you fetch 100 likely matches using an inexpensive algorithm like BM25, then have Jev score those 100 candidates for relevance against the original query.
我也一直在用它尝试搜索重排序：先用 BM25 等低成本算法获取 100 个高概率匹配结果，然后让 Jev 针对原始查询对这 100 个候选项的相关性进行打分。
Black boxes are back in fashion
黑盒模型再度流行
Something I've found a little uncomfortable about Jev is how it very much represents a regression even further towards black box machine learning systems.
我对 Jev 感到些许不安的一点是，它代表了机器学习系统进一步向黑盒化倒退的趋势。
LLMs are black boxes already - you can ask them to justify their decisions, but you can't guarantee that what they say is useful or accurate.
大语言模型（LLMs）本身就是黑盒——你可以要求它们解释自己的决策，但你无法保证它们的回答是有用或准确的。
Jev doesn't even give you that: put in all the text you want, the only thing you're going to get back is a floating point number. If Jev marks something as spam, which content signals tipped it off?
Jev 甚至不提供这种解释能力：你输入任意文本，唯一返回的是一个浮点数。如果 Jev 将某内容标记为垃圾信息，究竟是哪些内容信号触发了这一判断？
This also means that concerns about bias should be front and center. I really hope nobody uses Jev to rank job applicants - that floating point number could conceal all manner of unseen bias baked into the models, and experimentally picking that bias apart is going to be a tricky business.
这也意味着对偏见的担忧应当放在首位。我真的希望没有人使用 Jev 来对求职者进行排名——那个浮点数可能掩盖了模型中内嵌的各种隐性偏见，而通过实验手段剖析这些偏见将是一项棘手的工作。
(I tried one experiment where I had Jev score every city in the San Francisco Bay Area on a yes/no answer to whether they were a "Good city?" - it rated Cupertino top and East Palo Alto bottom. Huh.)
（我曾做过一个实验，让 Jev 对旧金山湾区的每个城市就“是好城市吗？”这一问题给出是/否评分——它将库比蒂诺评为最高，东帕洛阿尔托评为最低。真是令人惊讶。）
In practice, this all means that evals and structured experiments are even more important than they are for regular LLM projects. Thankfully, Jev is so cheap that running hundreds or even thousands of experimental prompts through it costs just a few cents.
在实践中，这意味着评估和结构化实验对于 Jev 而言，比对于常规 LLM 项目更为重要。值得庆幸的是，Jev 的成本极低，运行数百甚至数千个实验性提示词仅需几分钱。
Unconventional uses for Jev
Jev 的非传统用途
It's been really fun watching the wider community come up with potential use-cases for Jev over the past few days. Here are some creative ones that caught my eye:
过去几天里，看着更广泛的社区为 Jev 构思潜在用例，真的非常有趣。以下是一些引起我注意的创意用法：
- jevchat by Kyle Pena turns Jev into a (terrible) chat model. "At every step it asks Jev one question: Given the user's question and the reply written so far, which symbol comes next?". ericpruitt on Hacker News: "It's the digital equivalent of Morty speaking with the death crystal".
- jev-leftpad by Fatih Kadir Akın implements left-pad with the prompt "How many spaces are needed before value to reach targetLength?" and a choice query allowing options from "0 spaces are needed" to "10 spaces are needed".
- jev-2048 by Andy Gayton uses Jev to play the 2048 sliding puzzle game.
- Kyle Pena 开发的 jevchat 将 Jev 变成了一个（糟糕的）聊天模型。“它在每一步都会向 Jev 提出一个问题：鉴于用户的问题以及目前已生成的回复，下一个符号是什么？”。Hacker News 上的 ericpruitt 评论道：“这相当于 Morty 与死亡水晶对话的数字等价物。”
- Fatih Kadir Akın 开发的 jev-leftpad 使用提示词“需要多少个空格才能使值达到目标长度 targetLength？”来实现左填充功能，并通过选择式查询提供从“不需要空格”到“需要 10 个空格”的选项。
- Andy Gayton 开发的 jev-2048 利用 Jev 来玩 2048 滑动拼图游戏。
Open weight recreations
开放权重的复刻版本
There's also been a flurry of projects attempting to create a model like Jev using on top of open weight models. Kev is one interesting example, using Qwen 3.5 to produce 0.8B, 4B, and 9B models. Here's the accompanying Hacker News thread, where someone linked to a JevBench benchmark that has already cropped up to compare "Jev-class decision models".
近期也涌现出一批试图基于开源权重模型构建类似 Jev 模型的项目。Kev 是一个有趣的例子，它使用 Qwen 3.5 生成了 0.8B、4B 和 9B 规模的模型。以下是相关的 Hacker News 讨论帖，其中有人链接到了一个已出现的 JevBench 基准测试，用于比较“Jev 级决策模型”
Given Jev was released just under a week ago, the amount of activity around it is extremely impressive.
鉴于 Jev 发布至今不到一周，围绕它所产生的活动量令人印象深刻
Tags: ai, generative-ai, llms
标签：ai, generative-ai, llms
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力