The great master Li Mu and Kuaishou veteran Li Yan were exposed and switched to big models after leaving their jobs. ChatGPT set off a boom in AI entrepreneurship

Recently, Internet giants have joined the large model track one after another.

Yesterday, the news that Master Li Mu left Amazon to work as a model model exploded on everyone’s social networks like thunder.

Following this, it was revealed today that the new company established by former Kuaishou AI core boss Li Yan after leaving Kuaishou in 2022 also makes large models.

Obviously, since ChatGPT has allowed the world to see the application scenarios of AI, the competition among domestic enterprises in the AI ​​model layer has begun to intensify.

Kuaishou boss Li Yan started a business and entered multi-modality

Li Yan established the AI ​​company "Yuanshi Technology" in the second half of 2022, mainly engaged in multi-modal The development of large dynamic models.

Li Yan is an old employee of Kuaishou with a job number of around 75, and is also the core figure in the research and development of Kuaishou AI technology.

In November 2015, with the support of Su Hua, then CEO of Kuaishou, Li Yan established the first internal deep learning department DL (Deep Learning) group with the goal of building The algorithm model identifies video content that violates laws and regulations.

Subsequently, Kuaishou had more needs for video content understanding. In 2016, Li Yan changed the name of the team from the DL group to the MMU (Multimedia understanding, multimedia content understanding) group. In addition to solving security compliance issues, it also dabbled in the research and development of algorithm models in various forms such as voice, text, and images.

At the 2018 CNCC conference, Li Yan emphasized the importance of multimodal model technology in a speech titled "Multimodal Content Production and Understanding":

• Change the way of human-computer interaction
• Make information distribution more efficient

Take the short videos we often watch as an example, In addition to multi-modal information such as visual, auditory and text, user behavior is also another modal data.

In this way, the video itself and the user's behavior together constitute a very complex multi-modal problem.

The purpose of multimodal research is to make the way human-computer interaction becomes more and more natural and comfortable.

However, multi-modal research is quite difficult.

On the one hand, we must face the semantic gap problem of single modality and the heterogeneous gap problem of how to comprehensively model data of different modalities; on the other hand, we must also solve Missing data problem due to difficulty in constructing multimodal datasets.

At that time, many studies in the academic community still stayed in the single-modal field, but Li Yan firmly believed that multi-modality would become a more valuable research direction in the future.

His experience in Kuaishou gave Li Yan a deep understanding of the ecology of AI in short videos. In 2021, he chose to leave Kuaishou.

In the second half of 2022, he established Yuanshi Technology. According to 36Kr’s exclusive verification, Yuanshi Technology’s main focus is the research and development of multi-modal large models.

Master and disciple in one: write a book, start a business, start a business again

And yesterday, the news that Master Li Mu was suspected of joining a large model entrepreneurship was instantly posted on social networks Screen.

According to the public account "Dear Data", Alex Smola, the "father of parameter server", left Amazon in February this year and founded an artificial intelligence company called Boson.ai.

As for the introduction of this new company, there is not much information, and the official page is still under construction.

Link: https://boson.ai/

To be sure, we need to do large-scale model-related projects.

Also according to Alex’s LinkedIn profile, “We are doing something big. If you are interested in the scalable basic model, please contact me.”

#It is worth noting that on the company’s GitHub homepage, Amazon’s chief scientist Li Mu also contributed code.

# Therefore, it is speculated that Li Mu has joined Boson.ai and started a business with his mentor.

#However, so far, its homepage has not been updated.

Li Mu and Alex Smola founded a data analysis algorithm company called Marianas Labs in 2016.

#At that time, Li Mu served as CTO and co-founder.

Li Mu once mentioned in the article "The Five Years of Doctorate" that

At that time The popularity of deep learning has led to various large-scale acquisitions of start-up companies.

Alex worked with him for a long time with hundreds of thousands of angel investments. Alex wrote crawlers and he ran the model himself, and later sold it to a Small Public Company Company 1-Page.

The master and apprentice first met at Carnegie Mellon University (CMU) ).

In September 2012, Li Mu went to CMU for further studies, studying under Alex Smola.

#At that time, Alex was still working at Google and there was no funding, so they left him to Dave Andersen. Therefore, Li Mu had two mentors, one doing machine learning and the other doing distributed systems.

#In the first half of the year at CMU, Li Mu chatted with two mentors for an hour every week.

Because the two instructors have very different styles, and Alex reacts very quickly, it is difficult to keep up with his rhythm. If you want to explain your ideas, you need to do more homework.

And Dave will help Li Mu understand something thoroughly without giving many ideas.

# Under the guidance of two mentors, Li Mu grew up rapidly.

#In his second year of studying at CMU, while Yu Kai and others were doing deep learning, Li Mu also joined this research boom.

Based on his interest in distributed deep learning frameworks, he chose to cooperate with Chen Tianqi and use CXXNet as a starting point to do deep learning related projects.

When the two of them wrote the xgboost distributed startup script together, they discovered that file reading can be used by multiple projects.

In order to avoid reinventing the wheel, Li Mu and Chen Tianqi worked together to create an organization called DMLC on Github, and then created the DMLC, which became a great success. MXNet.

In July 2016, Alex joined Amazon. At the same time, Li Mu took MXNet to join Amazon as a part-time employee and chose to stay after graduation.

#During 2019, the master and apprentice also gave lectures together at UC Berkeley.

#In 2021, the two will also teach "Practical Machine Learning" together at Stanford University.

It is worth mentioning that the book "Hands-On Deep Learning" is Written by Li Mu, Aston Zhang, PhD in computer science at the University of Illinois at Urbana-Champaign, and his mentor Alex.

#This book has become very popular since its release. As one of the authors of MXNet, Li Mu's "Hands-On Deep Learning" is also written using the MXNet framework.

Who else is on the road to large models?

The multi-modal direction is what Li Yan has wanted to do for a long time. Li Mu followed his mentor to start a business, which may have been affected to some extent by the popularity of ChatGPT.

The competition among domestic enterprises in the AI ​​model layer has begun to intensify. The current large-scale model track is crowded with players from all walks of life, including the giants, big bosses, returnees/big factory executives, small startups transitioning, professors, and soy sauce factions.

On February 13, Wang Huiwen, who had retired from Meituan for two years, returned to the public eye with an "AI Hero List", saying that he would spend 50 million US dollars " Bring money to join the team", and "I don't care about the position, salary and title, I want to form a team."

## In the past, Wang Huiwen raised the ticket price for large-scale business startups to 50 million US dollars. Later, there was "Go out and ask Ask" founder Li Zhiwen officially announced the end of the large model competition.

Li Zhiwen led the team in 2020 to train the large model UCLAL

In addition, former Sogou CEO Wang Xiaochuan also issued a vague statement. Announced that he was about to enter the battlefield of "China's OpenAI" and admitted to 36Kr that he was making rapid preparations.

On February 26, Zhou Bowen, the founder and chief scientist of Xianyuan Technology, also released a message announcing the recruitment of partners. People, let’s work together to build the Chinese version of ChatGPT.

The recent surge in demand has shown that the potential market for domestically generated artificial intelligence products is surprisingly large.

The explosion of ChatGPT means that the singularity has arrived. It has triggered lower and deeper changes. The new generation of AI will integrate the physical world and the information world to realize knowledge and computing. , closed loop of reasoning.

In just two days, it was revealed that two big guys had quit their business to start a large model track. The press conferences predicted by domestic giants will be held within a few months.

Therefore, in this AI large model domestic pursuit competition that has been started since the beginning of the year, we may soon see some players sprint to the finish line.

The above is the detailed content of The great master Li Mu and Kuaishou veteran Li Yan were exposed and switched to big models after leaving their jobs. ChatGPT set off a boom in AI entrepreneurship. For more information, please follow other related articles on the PHP Chinese website!