Mobvoi's internal beta test explores the large model 'Sequence Monkey” to create exclusive CoPilot for individuals and enterprises

On April 20, Mobvoi held the 2023 AIGC strategy conference in Beijing with the theme of "AGI·Advent". At the meeting, Mobvoi announced an internal test to explore the large model "Sequence Monkey" and proposed that CoPilot will be everywhere. At the same time, based on the large model capabilities, it also launched a CoPilot product matrix for creators and an upgraded version for the C-side. Questions about voice assistant magic, as well as internal testing and exploration of enterprise-specific large-scale models for the B-side. Among them, the CoPilot product matrix for creators includes four AIGC products, namely the AI ​​writing platform "Qiaowen", the AI ​​painting platform "Yihua", the AI ​​dubbing platform "Magic Sound Workshop", and the digital human video and live broadcast platform "Wonderful Yuan".

Self-developed large model "Sequence Monkey" to help AGI "arrive"

Go out and ask about the large model "Sequence Monkey" It is a large language model with multi-modal generation capabilities. The model's language-centered capability system covers the six dimensions of "knowledge, dialogue, mathematics, logic, reasoning, and planning" and can support text generation, image generation, and 3D at the same time. Different tasks such as content generation, speech generation and speech recognition.

At present, the "Sequence Monkey" large model already has certain natural language understanding, knowledge, logic and reasoning capabilities. For more complex questions such as "Which provincial capital has more people, Hunan or Hubei?" Questions can give accurate results quickly.

According to reports, the inspiration for the Chinese name "Sequence Monkey" of Mobvoi's self-developed large model comes from the "Infinite Monkey" theorem proposed by mathematician Emile Borel in the early 20th century. According to this theorem, if a group of monkeys randomly hit a typewriter, they would eventually produce a complete work of Shakespeare. This theorem is based on probability theory and combinatorial mathematics and illustrates the unity of probability. The construction principle of Mobvoi's self-developed large model is similar to that of "Infinite Monkey". Massive text sequences are continuously trained by Mobvoi's independent algorithm and digested and understood by large-scale computing power, and become "Sequence Monkey".

Li Zhifei, founder and CEO of Mobvoi, said that large models are not just about large parameters. Today’s large models are in-depth modeling of Internet text, and Internet text is a mapping of world knowledge, so large models are a Cognitive model is a modeling of language. Language is the boundary of thinking, so large models have unlimited room for imagination. "Sequence Monkey" has already demonstrated its "emergent" ability during the training process. It is currently in the "enlightenment" stage and will improve faster and faster in the future. He also believes that being a human CoPilot will be the best "job" for large models, and CoPilot will be everywhere in the future.

One-stop CoPilot product matrix, opening up the entire content creation process

Based on the "Sequence Monkey" large model, Mobvoi has explored a variety of creator-oriented products during its internal testing AIGC products and applications build a one-stop CoPilot product matrix for thousands of creators, opening up the entire content creation process. Currently, Mobvoi has achieved the "trinity" of technology, products and commercialization, with over 5 million registered users and over one million paying users worldwide.

Qiaowen-Your AI Writing Assistant (write.mobvoi.com)

As the first CoPilot product unveiled at the press conference, the AI ​​of "Qiaowen" The writing ability covers the four major content creation scenarios of workplace, marketing, new media and creative writing, and can continuously provide users with inspiration and creative direction when writing year-end summaries, customer service words, scripts, advertising copywriting and other content. .

In order to better help users improve content creation efficiency, "Qiaowen" has opened eight major AI editing functions, namely style transformation, key point extraction, proofreading and error correction, continuation, rewriting, expansion, abbreviation, and translation. . "Qiaowen" can also automatically generate pictures during writing, providing users with a writing experience that includes both pictures and text.

Yuhua-your AI painting assistant (paint.mobvoi.com)

Faced with designers, illustrators, etc. who have a strong demand for design in addition to text The creators of Mobvoi.com explored the "Yihua" AI painting platform in internal testing. "Word Painting" supports 8 creative styles including two-dimensional, steampunk, and illustration. Users only need to input text, and "Yihuahua" can generate 8 2K high-resolution images with realistic light and shadow and rich details at one time.

In addition to using text to create pictures, it also has AI drawing capabilities such as drawings to create pictures, animation generation, and personalized avatar generation, which greatly enriches users' creative methods. For enterprise users, "Yihua" also supports exclusive model customization, allowing enterprise users to customize the model style according to their own needs, and supports multi-person collaborative production to better meet the drawing needs of enterprises.

Currently, "Yiyanhua" has reached an exploration intention with the home decoration design platform Kujiale. On the Kujiale platform, users can use "Yiyanhua" to describe their needs in words. You can easily change the decoration style, adjust the position of furniture, etc., and then design your favorite decoration plan.

Moyin Workshop-Your AI dubbing assistant (moyin.com)

For AI dubbing scenarios, Mobvoi has launched a new generation of AI dubbing product "Magic Sound Workshop". "Magic Sound Workshop" (overseas version of DupDub) is the world's leading full-process one-stop AI dubbing platform, which has opened to users more than 1,000 timbres, more than 2,000 voice styles, and more than 20 dialects and foreign languages.

With the support of large model technology, "Magic Sound Workshop" is the world's first dubbing platform equipped with large model AI writing functions, covering multiple scenarios such as AI writing, AI dubbing and editing. With its assistance, users can easily complete content creation that integrates copywriting and dubbing, such as film and television commentary, audio books, online education, and news broadcasts. At present, "Magic Sound Workshop" has reached cooperation with leading companies in many industries such as WeChat Reading, Juvenile Get, and Volkswagen.

In order to provide users with a better dubbing experience, "Magic Sound Workshop" supports the adjustment of 7 emotions including calm, sad, and happy for the selected voice. 10 kinds of characters, including 10 characters, etc., have been migrated. At the same time, AI sound personalized editing functions such as rhythm adjustment, local speed change, and multi-person dubbing have been opened, allowing users to edit sounds like editing documents with Word.

In addition to selecting and editing sounds, the "Magic Sound Workshop" will also launch a "pinch sound" function based on large-scale model language generation capabilities, allowing users to freely select gender, age, language, style and emotion, etc. Sound characteristics, create your favorite sound from 0 to 1.

Wonderful Yuan-Your AI Digital Avatar (weta365.com)

Based on the ability to generate text, images, sounds, etc., in order to make video content creators more To quickly and better create novel, lively and interesting works, Mobvoi’s internal beta test explored the AI ​​digital image creation and live broadcast platform “Wonderful Yuan”.

According to reports, the "Wonderful Yuan" platform currently has over 100 digital people, over 1,000 3D digital assets, and over 1,000 sounds. With multi-modal generation technology, the "Wonderful Yuan" platform currently supports three different forms of digital human generation: picture modeling (2D digital human), video modeling (2.5D digital human), and 3D modeling (3D digital human). . Its image cloning function only requires a 5-minute live video material to reproduce the user's image and demeanor 1:1, creating a digital clone for the user with consistent voice and natural movements.

From individuals to enterprises, CoPilot will be everywhere

Mobvoi has a deep accumulation in the field of voice assistants, and its research on AI voice can be traced back to ten years Previously, Mobvoi released its first voice assistant "Xiaowen Assistant" in 2014, iterated on "Xiaowen Secretary" in 2015, launched full-scenario VPA in 2017, and upgraded VPA in 2019. After ten years of accumulation and development, CoPilot-Magic Xiaowen, which was explored in internal testing by Mobvoi at this conference, can be understood as a synthesis of Siri and ChatGPT.

"Hello, hello, where is the nearest hot pot restaurant?" ""Hello, can you tell me the weather in Beijing tomorrow?" People are accustomed to looking up relevant information before necessitating food, clothing, housing, and transportation. As Develop an intelligent voice assistant for individual users. "Magic Question" can help users check encyclopedia information, weather, restaurants, and hotels, and can also talk to users freely, allowing users to obtain information more conveniently and quickly.

According to Mobvoi, “CoPilot” is derived from the concept of a super assistant. It will have a high-intelligence brain based on a large model, which can analyze massive data and communicate with humans instantly and accurately; it will also have beautiful voices, beautiful The image can be equipped with any hardware platform, such as mobile phones, watches, car machines, etc.; it can also be adapted to various industries and transform into high-quality teaching teachers, knowledgeable lawyers, professional doctors, financial management customer service, etc., when needed time, demonstrate professional knowledge anytime and anywhere, and share the pressure for others.

"Sequence Monkey" opens up the ecosystem and empowers more industries

Based on the "Sequence Monkey" large model For bottom-level capabilities, for B-end users and vertical fields, "Mobvoi Copilot" not only provides general capability support services and digital human image customization services, but will also open up its own role capabilities and continue to iterate data interfaces. Enterprise users in need can log in to the web to call API services, and train on industry-specific content by uploading documents to achieve customized voice interaction. The URL of Sequence Monkey is openapi.mobvoi.com.

Currently, Mobvoi has reached cooperation with the first batch of internal beta exploration partners in the top ten industries, including automobiles, education, law firms, finance, medical care, tourism, etc. In the future, Mobvoi “CoPilot” will Gradually empower more industries, help more companies have their own exclusive large models, and create their own exclusive "CoPilot".

The above is the detailed content of Mobvoi's internal beta test explores the large model 'Sequence Monkey” to create exclusive CoPilot for individuals and enterprises. For more information, please follow other related articles on the PHP Chinese website!