Yang Jing Fengse comes from the Temple of Concave Fei
Qubits | Official account QbitAI
Another AI bull, plunged into the wave of big models.
Qubit has recently learned that a new AI company is brewing and targeting the direction of AI Infra (AI architecture).
Even if it keeps a low profile, it is hot underwater and is highly favored:
It is reported that the first round of financing intentions has been successfully obtained.
So what kind of people dare to start a business from the framework? And what kind of project can be so optimistic at birth?
Inseparable from those who appeared on the list of founding teams -
Frame god Jia Yangqing.
As one of the most powerful Chinese in the field of AI architecture, he founded the well-known deep learning framework Caffe, and is also a core member and co-leader of TensorFlow and Pytorch.
In addition, he is said to have been working with his comrades for many years, who have worked together on ONNX, PyTorch and Caffe2 projects.
It is said that the past week has been the most explosive week in the process of AI and even human science and technology.
But now it seems that everything is just beginning.
Because including Jia Yangqing, Li Mu, Li Kaifu and other AI bulls are now at the forefront of the wave.
What does the new company do?
Qubit learned that the direction of Jia Yangqing's plan to start a business this time belongs to an AI infrastructure provider, focusing on AI applications including large models.
Specifically, it's about creating a unified, multi-cloud, and observable AI platform.
The goal is to improve the efficiency of AI productization, including developer efficiency and system efficiency. End-to-end solutions to help enterprises and developers deploy AI efficiently.
For example, standardized cloud IaaS (infrastructure as a service) platform, modular AI algorithm lifecycle management platform, etc.
In fact, this is evident from his previous work and public speeches.
He founded the Caffe deep learning framework during his Ph.D. at Berkeley and is a core member of TensorFlow and co-leader of PyTorch and ONNX.
At Alibaba, he led the team to upgrade big data and artificial intelligence platform products represented by ODPS and PAI, supporting the needs of the entire group for big data storage, computing, analysis and machine learning. Support for the release of the ModelScope open source model community.
At the MEET2023 Intelligent Future Conference, he talked about the two keys to the efficient application and deployment of artificial intelligence -
AI engineering and open source.
Engineering, so that the path of development and iteration to application becomes more efficient; Open source can make work faster and achieve a win-win situation in the market.
On this basis, he also talked about four trends in AI industry applications:
Cloud-native AI engineering platform; End-to-end optimized heterogeneous computing system; Through the combination of systems, intelligent products close to users and open source algorithms are used to help the wide application of AI in vertical industries.
Now these trends are also reflected in entrepreneurial projects.
It is said that working with him to start a business is also Jia Yangqing's long-time comrade-in-arms, who has been a colleague during Facebook's period, once the head of Facebook AI Infra technology, and then co-participated in the development of ONNX, PyTorch and Caffe2, and was the director of Ali's artificial intelligence platform. and a chief engineer from Uber, who has worked at Meta and Ali.
The most powerful Chinese in the field of AI architecture (none of them)
Jia Yangqing is a native of Shaoxing, Zhejiang. His parents were both middle school Chinese teachers, but he became interested in programming in the third year of junior high school and began to study.
Tsinghua was ranked in the college entrance examination, but unfortunately missed the computer major, so he entered the automation major. He then went on to complete a master's degree in control science and engineering at Tsinghua University, and then went to UC Berkeley to pursue a Ph.D. in computer science.
That is, during his doctoral studies, Jia Yangqing completed his famous work:
Open source general learning framework Caffe.
Caffe was inspired by Cuda-Convnet, but its design is more systematic, more abstract, and has full unit testing. Once born, it has become the most popular tool for many deep learning beginners, and has also been adopted by Microsoft, Yahoo, NVIDIA, Adobe and other large companies.
There is also an interesting story about the development process of Caffe.
It is said that Jia Yangqing, who was still a doctor at the time, was actually very "picky" when buying equipment, relying on a K20 GPU donated by NVIDIA, plus a computer of more than 600 yuan saved on Amazon.
According to him, another important reason for developing Caffe was not wanting to write a doctoral dissertation.
And the name "Caffe" for this framework is also because I drank too much coffee during my internship at Google and wanted to quit.
Because Caffe's development work took up too much of his time, his graduation thesis, which was supposed to be completed in December of that year, was delayed until the next year.
But the mentor has always supported his approach:
Do you want to spend more time writing a graduation thesis that you don't expect to care about, or do you want to spend more time writing a system that everyone will use in the future?
Everyone knows the story that followed.
Shortly after Caffe's completion, Jia Yangqing also successfully graduated with a PhD and immediately joined Google Brain, serving as a research scientist under Jeff Dean, and starting another world-class masterpiece: as a core author, building TensorFlow.
Two years later, Jia Yangqing jumped to Facebook and continued to make all-out efforts in the field of AI architecture:
First co-developed the world's first deep learning engine, Caffe2Go, deployed on more than 1 billion mobile phones, and then soon launched Caffe2.
Then, the first version of ONNX, an intermediate machine learning framework he pioneered, was released.
The following year, on the basis of Caffe2 and PyTorch 0.4, and the integration of ONNX format, PyTorch 1.0 was born, and Jia Yangqing was the co-leader of the project.
Before Jia Yangqing, few people in the world completed the development of such a number of industry-heavy architectures. In the field of Chinese, it is unprecedented.
It can be called the first Chinese in the field of AI framework.
Of course, to complete this series of fruits, Jia Yangqing is already the director of AI architecture at Facebook (now Meta).
But in 2019, Jia Yangqing made an important decision: returned to his hometown Zhejiang, joined Alibaba, became the vice president of Ali technology at the age of 35, and after joining the company, Jia Yangqing mainly led the research and development of the big data computing platform.
At that time, Professor Chen Yiran of Duke University commented that Jia Yangqing's joining allowed Ali to "open the 'one hard and one soft' two-university overlord leadership model, and it is a breakthrough in the field of AI research."
The tide beats the shore: AI bulls can't sit still
Yesterday, Kai-Fu Lee, CEO of Innovation Works, announced that the new company, Project AI 2.0, officially started team formation around the platform and productivity application ecology in the AI 2.0 era, and recruited talents from large models, NLP, multimodal, algorithms to distributed architecture/Infrastucture and other directions around the world.
Two weeks ago, Amazon's chief scientist Li Mu also announced his departure to join forces with his mentor to start Boson.ai, specializing in large models for everyone.
Globally, Google, Deepmind and other engineers left to start businesses in multiple directions around large models.
A wave of startups driven by ChatGPT is unstoppable.
— End —
Qubits QbitAI · Headline number signed
Follow us and be the first to know the latest scientific and technological trends