如何定义B端产品及B端产品经理方法论
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
Amu Chat AI (Intelligent Agent)
follow with interest
2024-07-25
1 Comment
8967 views
15 collections
14 minutes
The irreplaceable value of a product manager is the ability to accurately identify and meet user needs, translate those needs into products, coordinate resources to drive product implementation, and create business value.
In depth exploration of Kimi's unique features, target audience, functional highlights, and how to efficiently utilize this AI assistant to unlock its potential applications in various fields, leading readers into a new era of smarter and more efficient work and learning.
1、 Kimi Introduction
Kimi Chat is an intelligent assistant product launched by Moon Dark Side on October 9, 2023. It innovatively supports the input of 200000 Chinese characters, becoming the longest contextual input supported by productized large model services in the global market.
In addition, on March 18, 2024, Moonlit announced that its self-developed Kimi intelligent assistant can support 2 million words of ultra long lossless context and has started internal testing of the product.
According to actual testing, the maximum context length of ChatGPT 4 can reach 32768 tokens (theoretically supporting approximately 32768 Chinese characters), while the maximum context length of Doubao can reach 35000 Chinese characters.
It should be noted that although AI products can support longer contextual inputs, factors such as computational complexity and response speed still need to be considered in practical applications.
How to define B-end products and B-end product manager methodology
Compared to C-end products, the biggest feature of B-end products is that they are aimed at specific domain users and have a much smaller quantity, but they pay more attention to in-depth exploration of user professional domain operation processes - that is, they have stronger professionalism and are more closely integrated with business.
View details>
An excessively long input sequence may prolong the processing time of the model, thereby affecting the user experience.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
1. Target audience
Academic research population: Quickly read and deeply understand a large amount of literature, grasp the essence of literature in their native language, explain complex academic concepts, analyze research results, write papers, and respond to reviewers' suggestions.
Internet practitioners: collect information efficiently, assist competitive product analysis, operation planning and other scheme writing, and provide insight.
Programmer: Assisted in programming, problem solving, code commenting, API documentation reading, supports multiple programming languages such as Python, C++, Java, etc., helping you learn programming and solve technical problems.
We media and content creators: learn the style of specific official account, documents, and web links to assist in creation; Quickly gather the necessary information for creation, provide rich materials and inspiration.
Legal professionals: efficiently handle a large amount of case materials, organize evidence, sort out legal policies, assist you in writing legal documents and reports, and improve work efficiency.
2. Feature Highlights
(1) Online search: There is a deadline for the training data of large models, but there is no such restriction for online search. Kimi supports searching for "real-time" information, quickly integrating and providing detailed responses, while also providing sources of information, making conversations both rich and accurate.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
The reason why "real-time" is quoted here is that Kimi's real-time is relative to the existing content on the Internet. If no one has published content similar to your search, Kimi cannot give an answer.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
(2) Efficient Reading: Provide Kimi with a link that allows for quick summarization, whether from domestic or international sources, in both Chinese and English. This will help you accurately understand literature, lengthy reports, complex contracts, and lengthy web pages, and provide profound insights and analysis.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
(3) Professional interpretation documents: Whether it's financial analysis, legal consulting, or market research, Kimi can provide support with professional expertise. For long documents and multiple files, it can quickly summarize, translate, and answer questions. It supports uploading PDF, Word, Excel, PPT, TXT files, and images.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
(4) Organize information: If you have stomach discomfort and go to the hospital to get medicine, send the invoice to Kimi, which can help you organize the content you need. Even for complex and lengthy meeting minutes, Kimi can intelligently recognize and organize them, automatically extract key points, making data organization no longer complicated and information clear and concise.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
(5) Assisted Creation: Based on the webpage links, files, and instructions provided by users, Kimi can assist you in organizing outlines, continuing articles, creating copy, writing weekly reports, and developing plans, becoming a powerful assistant for your content creation, providing endless inspiration.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
(6) Programming Assistant: Kimi can help you quickly read API documentation, locate required information, and generate code and debug quickly according to your needs, covering multiple programming languages such as Python, C++, Java, etc.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
2、 How individual users use Kimi
According to the AI Product List (aicpb. com), in April 2024, the web version of Kimi Intelligent Assistant had a traffic volume of 20.04 million. According to data from Quest Mobile, in March 2024, the monthly active users of the Kimi Smart Assistant app were 5.897 million, and the monthly active users of the Kimi Smart Assistant WeChat mini program have exceeded 911000.
1. Web version
Users can access kimi.moonshot.cn through a browser to use the web version of Kimi. After entering the webpage, you need to register an account first in order to use the conversation function. After logging in, the user enters a question or requirement in the input box, and Kimi will provide answers and services for you.
2. Browser plugin
The Kimi browser plugin supports mainstream browsers such as Google Chrome, Firefox, Microsoft Edge, Safari, Opera, and has features such as instant Q&A, full-text summarization, and underlined interaction.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
3. Mobile APP
The Kimi mobile app will be launched on October 9, 2023, open for internal testing on October 10, and open for service to the whole society on November 16. In the past month's updates, Kimi has added the function of adjusting the speed of voice playback, editing the latest sent message, optimizing the voice recognition experience and interface details, and allowing the browser within the app to forward links.
4. WeChat Mini Program
The smoothness of mini programs is usually good, but there may be some lag when switching interface levels or loading dynamic data, and the interface functions are relatively simple. It is recommended to use mobile apps.
3、 Kimi Price
1. Model inference
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
Here 1M=1000000, the prices in the table represent the price per 1M tokens consumed. The difference between the above models lies in their maximum context length, which includes both the input message and the generated output, and there is no significant difference in their effectiveness. Among them, the moonhot-v1-8k model supports approximately 15200 Chinese characters, the moonhot-v1-32k model supports approximately 60000 Chinese characters, and the moonhot-v1-128k model supports approximately 200000 Chinese characters.
The Chat Completion interface is charged based on the actual consumption of input and output tokens, while the file related interface (file content extraction/file storage) is free for a limited time.
2. Context caching
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
The price is in the public beta period and may be adjusted at any time.
Regarding billing logic, Cache resource fee=Cache creation fee+Cache storage fee. A Chat call that uses Cache once is charged=Cache call fee+Input Tokens fee for Chat that does not match Cache+Output Tokens fee.
3. Recharge and speed limit
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
In order to ensure fairness in overall resource allocation and prevent malicious attacks, the dark side of the month will impose rate limits based on the cumulative recharge amount of the account. Rate limiting is a common practice in API interfaces, with the following main considerations:
Helps prevent the abuse or misuse of APIs. For example, malicious actors may flood APIs with a large number of requests, attempting to overload them or cause service interruptions. By setting speed limits, such behavior can be prevented.
Rate limiting helps ensure that everyone can access the API fairly. If a person or organization makes too many requests, it may slow down everyone's API. By limiting the number of requests that a single user can make, as many people as possible have the opportunity to use the API without encountering speed issues.
Rate limiting can help manage the total load of the cluster. If there is a sharp increase in API requests, it may put pressure on the server and lead to performance issues. Setting speed limits can help maintain a smooth and consistent experience for all users.
4、 Kimi+App Market
Kimi+is a feature provided by the Dark Side of the Moon, allowing users to customize Kimi models according to specific needs. Users can summon specific Kimi+for interaction by entering the "@" symbol in the dialog box. Kimi+is more specialized and targeted in certain specific fields compared to Kimi. Here are some Kimi+features and advantages:
Professional domain knowledge: Kimi+can provide deeply customized services based on different professional fields, such as medicine, law, programming, language translation, etc.
Language proficiency: Kimi+can focus on translating and communicating in specific languages, providing more accurate language services.
Programming Assistance: For programmers, Kimi+can provide services such as code review, programming advice, and technical problem-solving.
Academic research: Kimi+can assist users in conducting academic research activities such as literature search, data analysis, and academic writing.
Educational tutoring: Kimi+can provide students with subject tutoring, homework assistance, exam preparation, and other services.
Personalized service: Users can customize Kimi+according to their own needs and receive services that better meet their individual needs.
Multi tasking: Kimi+can handle multiple tasks simultaneously, improving work efficiency.
Real time update: Kimi+can constantly update according to the latest knowledge and technology to maintain the progressiveness and accuracy of the service.
Comprehensive Perspective on Kimi: Features, Versions, Prices, Advantages, Applications
5、 Potential issues and precautions when using Kimi
Clarifying some potential issues and precautions when using Kimi can help you make more effective use of this service while avoiding potential challenges. Here are several key points:
Language limitations: Although Kimi supports conversations in multiple languages, she is primarily proficient in Chinese and English, and may have certain limitations in supporting other languages.
Content restriction: Kimi follows the laws of the People's Republic of China and refuses to answer questions related to terrorism, racial discrimination, yellow violence, political sensitivity, etc.
Privacy protection: When using Kimi, do not share sensitive personal information, such as ID number, bank account, etc.
File processing capability: Kimi can handle multiple file formats (such as TXT, PDF, Word documents, etc.), but there is a limit on the total word count (not exceeding 200000 words).
Search ability: Kimi has search ability, but the accuracy and reliability of search results depend on the source of the search results.
File generation restriction: Kimi cannot generate files or provide download links, only text replies can be provided.
Real time: Kimi's response is real-time, but for some complex issues, it may take some time to process.
Technical limitations: Kimi may not be able to provide in-depth answers to certain technical issues, especially those that require specialized domain knowledge.
Usage scenario: Kimi is more suitable for processing textual information and providing auxiliary advice, rather than replacing expert opinions in professional fields.
This article was originally published by @ Amu Chat AI (Intelligent Agent) on Everyone is a Product Manager. Reproduction without the author's permission is prohibited
Screenshot from Kimi's official website
The viewpoint of this article only represents the author himself, and everyone is a product manager. The platform only provides information storage space services
深入探讨Kimi的独特之处、适用人群、功能亮点以及如何高效利用这款AI助手，解锁其在各个领域的应用潜力，引领读者走进一个更智能、高效的工作与学习新时代。
Kimi Chat 是月之暗面于 2023 年 10月 9 日推出的智能助手产品，它开创性地支持输入 20万汉字，成为全球市场中可产品化使用的大模型服务里所能支持的最长上下文输入。
此外，在 2024 年 3 月 18 日，月之暗面宣告其自研的 kimi 智能助手已能支持 200万字的超长无损上下文，并开启了产品内测。
据实测，ChatGPT 4 的上下文最大长度可达 32768 个 token（理论上约能支持 32768 个汉字），而豆包的上下文最大长度能达 3.5 万汉字。
需要留意的是，尽管 AI 产品能够支持较长的上下文输入，然而在实际运用时，仍需斟酌计算复杂度与响应速度等要素。
过长的输入序列或许会致使模型处理时间延长，进而影响使用体验。
学术科研人群：快速阅读并深入理解大量文献，用母语掌握文献的精髓，解释复杂学术概念，分析研究结果，撰写论文，回应审稿人的建议。
互联网从业者：高效搜集信息，辅助竞品分析、运营策划等方案撰写，并提供洞察。
程序员：辅助编程、问题解答、代码注释、API 文档阅读，支持 Python、C++、Java 等多种编程语言，帮助你学习编程、解决技术问题。
自媒体与内容创作者：学习特定公众号、文件、网页链接的风格，辅助创作；快速搜集创作所需信息，提供丰富的资料与灵感。
法律从业人员：高效处理大量案件资料、整理证据、梳理法律政策，协助你撰写法律文书和报告，提升工作效率。
（1）联网搜索：大模型的训练数据存在截止日期，联网搜索则无此限制。Kimi 支持搜索“实时”信息，能迅速整合并给出详尽答复，同时提供信息来源，使对话既丰富又准确。
这里之所以给“实时”加上引号，是因为 Kimi 的实时性是相对于互联网中已有的内容而言，如果无人发表过与您搜索类似的内容，Kimi 也无法给出回答。
（2） 高效阅读：给 Kimi 一个链接，不论其来自国内外，不论中英文，它都能迅速进行摘要，助您精准理解文献书籍、长篇报告、复杂合同、网页长文，并提供深刻的洞察与分析。
（3） 专业解读文件：不管是金融分析、法律咨询，还是市场调研，Kimi 都能以专业水平提供支持。对于超长文档和多个文件，都能够快速摘要、翻译、答疑，支持上传 PDF、Word、Excel、PPT、TXT 文件和图片。
（4）整理资料：倘若胃不舒服去医院拿了药，将发票传给 Kimi，它能帮您整理出您所需的内容。即便是复杂冗长的会议记录，Kimi 也能智能识别并整理，自动提取关键要点，让资料整理不再繁杂，信息清晰明了。
（5）辅助创作：依据用户提供的网页链接、文件、指令，Kimi 能够协助您梳理大纲、续写文章、创作文案、撰写周报、制定方案，成为您内容创作的得力助手，提供无尽的灵感。
（6）编程助手：Kimi 能够帮助您迅速阅读 API 文档，定位所需信息，还能根据您的需求生成代码、快速 Debug，涵盖 Python、C++、Java 等多种编程语言。
据 ai 产品榜（aicpb.com）的数据，2024 年 4 月，kimi 智能助手网页版的访问量达到 2004 万。而 questmobile 的数据显示，2024 年 3 月，kimi 智能助手 app 的月活跃用户量为 589.7 万，kimi 智能助手微信小程序的月活跃用户已突破 91.1 万。
用户能够通过浏览器访问 kimi.moonshot.cn 来使用网页版 Kimi。进入网页后，需先注册账号，方能使用对话功能。登录后，用户在输入框中输入问题或需求，Kimi 便会为您提供回答与服务。
Kimi 浏览器插件支持 Google Chrome、Firefox、Microsoft Edge、Safari、Opera 等主流浏览器，具备即时问答、全文摘要和划线互动等功能。
Kimi 手机 APP 于 2023 年 10月 9 日推出，10月 10日开放内测，11 月 16 日面向全社会开放服务。在近一个月的更新中，Kimi 新增了语音播放调整倍速的功能，可编辑最新一条发送过的消息，优化了语音识别体验和界面细节，并且 APP 内浏览器能够转发链接。
小程序的流畅度通常较好，但在切换界面层级较深或加载动态数据时，可能会出现部分卡顿的情况，而且界面功能相对较为单一，建议使用手机 APP。
此处 1M = 1,000,000，表格中的价格代表每消耗 1M tokens 的价格。以上模型的区别在于它们的最大上下文长度，这个长度包括了输入消息和生成的输出，在效果上并没有什么区别。其中，moonshot-v1-8k 模型大约支持1.52万个汉字，moonshot-v1-32k 模型大约支持6万个汉字，moonshot-v1-128k 模型大约支持20万个汉字。
Chat Completion 接口收费则按照实际输入输出 tokens 的消耗计费，文件相关接口（文件内容抽取/文件存储）接口限时免费。
价格处于公测期，可能随时调整。
关于计费逻辑，Cache 资源收费= Cache 创建收费+ Cache 存储收费，一次使用了 Cache 的 Chat 调用收费= Cache 调用收费+ Chat 未匹配 Cache 的 Input Tokens 收费+ Output Tokens 收费。
为了整体资源分配的公平性，同时防止恶意攻击，月之暗面将基于账户的累计充值金额进行速率限制。速率限制是API接口的常见做法，主要有以下几个考量：
有助于防止滥用或误用API。例如，恶意行为者可能会通过大量请求来淹没API，试图使其过载或导致服务中断。通过设置速率限制，可以防范这样的行为。
速率限制有助于确保每个人都能公平地访问API。如果一个人或组织发出过多的请求，可能会拖慢所有人的API。通过限制单个用户可以发出的请求数量，那么尽可能多的人有机会使用API而不会遇到速度减慢的问题。
速率限制可以帮助管理集群总负载。如果对API的请求急剧增加，可能会给服务器带来压力并导致性能问题。通过设置速率限制将可以帮助为所有用户维护一个平稳且一致的体验。
Kimi+是月之暗面提供的一项功能，允许用户根据特定需求自定义 Kimi 模型。用户在对话框中输入“@”符号，便可召唤特定的 Kimi+进行交互。Kimi+相比 Kimi 在某些特定领域更具专业性和针对性。以下是一些 Kimi+功能和优势：
专业领域知识：Kimi+能够依据不同的专业领域提供深度定制的服务，例如医学、法律、编程、语言翻译等。
语言能力：Kimi+可以专注于特定语言的翻译和交流，提供更准确的语言服务。
编程辅助：对于程序员来说，Kimi+可以提供代码审查、编程建议、技术问题解答等服务。
学术研究：Kimi+可以帮助用户进行文献检索、数据分析、学术写作等学术研究活动。
教育辅导：Kimi+可以为学生提供学科辅导、作业帮助、考试准备等服务。
个性化服务：用户可以根据自己的需求定制Kimi+，获得更符合个人需求的服务。
多任务处理：Kimi+可以同时处理多个任务，提高工作效率。
实时更新：Kimi+可以根据最新的知识和技术不断更新，保持服务的先进性和准确性。
在使用 Kimi 时，明晰一些潜在的问题和注意事项，能够助您更有效地利用此项服务，同时规避可能遭遇的难题。以下是若干关键要点：
语言限制：尽管 Kimi 支持多种语言的对话，但主要擅长中文和英文，对其他语言的支持可能存在一定局限。
内容限制：Kimi 遵循中华人民共和国法律，拒绝回答涉及恐怖主义、种族歧视、黄色暴力、政治敏感等问题。
隐私保护：使用 Kimi 时，切勿分享敏感的个人信息，例如身份证号、银行账户等。
文件处理能力：Kimi 能够处理多种文件格式（如 TXT、PDF、Word 文档等），但总字数存在限制（不超过 20万字）。
搜索能力：Kimi 具备搜索能力，然而搜索结果的准确性和可靠性取决于搜索结果的来源。
文件生成限制：Kimi 无法生成文件或提供下载链接，仅能提供文字回复。
实时性：Kimi 的回复是实时的，但对于一些复杂问题，可能需要一些时间来处理。
技术限制：Kimi 在某些技术问题上可能无法提供深入的解答，特别是那些需要专业领域知识的问题。
使用场景：Kimi 更适合处理文本信息和提供辅助性建议，而不是替代专业领域的专家意见。
本文由 @阿木聊AI（智能体） 原创发布于人人都是产品经理。未经作者许可，禁止转载
题图来自Kimi官网截图
该文观点仅代表作者本人，人人都是产品经理平台仅提供信息存储空间服务