Loss of 53 billion! Massive layoffs...

Recently, there have been rumors that SenseTime is undergoing a new round of organizational restructuring and layoffs. In response, SenseTime said that the company is actively promoting strategic transformation, focusing on key businesses and strategic growth areas of "large devices-large models-applications", and optimizing and adjusting the organizational and talent structures accordingly to better meet business development needs. At present, SenseTime's social recruitment and campus recruitment are proceeding normally, and the overall business is developing steadily.

On October 22, Xu Li, Chairman and CEO of SenseTime, published an internal letter titled "SenseTime's 10th Anniversary: ​​Focus, Integration of Knowledge and Action", looking forward to AGI (general artificial intelligence) and the next decade. In it, Xu Li mentioned that at the organizational level, around strategy and core resources, a more centralized and efficient organizational structure will be built to promote the concentration and intensive investment of resources and accelerate the process of rejuvenation of organization and management.

According to the Science and Technology Innovation Board Daily, after the release of the letter to all employees, SenseTime launched organizational adjustments on the same day, involving layoffs. The company focused on retaining the business related to large models and large devices, and the rest of the business lines including security, autonomous driving, and medical care were adjusted. According to Digital Intelligence Frontline, layoffs are still in progress, "Shanghai has vacated a large conference room", the scale of layoffs is unknown, and the compensation plan is N+1 .

In the letter to all employees, Xu Li also talked about the milestones in the field of artificial intelligence (AI). Both the Nobel Prize in Physics and the Nobel Prize in Chemistry were awarded to researchers related to AI. He believed and looked forward to the arrival of the era of general artificial intelligence. He also compared traditional AI 1.0 and generative large model AI 2.0. The important difference lies in the change in AI cost structure.

Xu Li said that in the era of generative large-model AI 2.0, the cost of model production mainly lies in the investment in computing resources. As the law of scaling is verified in large language models, multimodal models, video generation models, and slow thinking and reasoning processes, the cost of producing and using large models can be directly equivalent to the consumption of computing resources. In short, the popularization and commercialization of generative large-model AI must be committed to reducing the production and use costs of large models, which will inevitably require the iteration and optimization of computing power in combination with large models, and it is also necessary to iterate the design and application of large models according to the characteristics of computing resources.

The letter to all employees pointed out that in the field of generative large-model AI, SenseTime's core strategy is to achieve seamless integration of large computing devices (SCO), large models, and applications (CNI), with applications driving models and models driving optimization of computing power. The company has established a trinity strategy of "large devices-large models-applications", aiming to improve the efficiency of computing resources and serve customers well through order-of-magnitude optimization. SenseTime is committed to becoming the largest model service provider that understands computing power the most, and the computing power service provider that understands large models the most.

According to official data, SenseTime is an artificial intelligence software company with the mission of "sticking to originality and letting AI lead human progress". It aims to conduct cutting-edge research in artificial intelligence and create a more scalable and inclusive artificial intelligence software platform. SenseTime has long been invested in original technology research, continuously enhancing multi-modal and multi-task general artificial intelligence capabilities, covering key technology areas such as perceptual intelligence, natural language processing, decision-making intelligence, and intelligent content generation, as well as key capabilities including AI chips, AI sensors, and AI computing infrastructure. Previously, SenseTime was listed on the main board of the Hong Kong Stock Exchange.

According to the public financial report, in the first half of 2024, SenseTime achieved total revenue of 1.739 billion yuan, a year-on-year increase of 21.4%. Among the three major businesses, generative AI revenue was 1.051 billion yuan, a year-on-year increase of 255.7%, accounting for 60% of the total revenue, surpassing the traditional AI business for the first time in history and becoming the company's dominant business.

According to the financial reports of previous years, SenseTime's net profit from 2018 to 2023 was -3.433 billion yuan, -4.968 billion yuan, -12.158 billion yuan, -17.177 billion yuan, -6.093 billion yuan and -6.495 billion yuan respectively, with a total loss of 50.324 billion yuan in these six years. Together with the loss of 2.457 billion yuan in the first half of this year, the cumulative loss is as high as 52.8 billion yuan.

The following is a letter from all employees of SenseTime on its 10th anniversary:

Dear SenseTime students:

Hello everyone! SenseTime celebrated its tenth birthday this month. At the same time, the AI ​​community also reached a major milestone: the Nobel Prize in Physics recognized the work of "using physics to promote AI", while the Nobel Prize in Chemistry recognized the achievement of "using AI to promote protein structure prediction". This marks the interaction between AI and science, which are both research purposes and research tools for each other, and heralds the opening of a new paradigm in a series of disciplines.

The new £50 note features a quote beneath Alan Turing's portrait: "This is but a foreshadowing of things to come, and a microcosm of things to come".

“This is only a foretaste of what is to come and only the shadow of what is going to be.” - This may be the best commentary on our era.

We believe in and look forward to the arrival of the era of general artificial intelligence. Pragmatically speaking, we have taken two steps: the so-called traditional AI 1.0 and generative large model AI 2.0. Usually, AI 1.0 is regarded as a special-purpose intelligence, focusing on single tasks and information processing; AI 2.0 is regarded as a general intelligence, emphasizing multi-task and content generation. However, although such a description is simple and easy to understand, it is not completely accurate, because there is no clear boundary between general and special, and the implementation of AI must be scenario-based, such as the application of generative large models in vertical fields. In our view, an important difference between AI 1.0 and AI 2.0 is the change in the cost structure of AI.

In the traditional AI 1.0 era, the main cost of model production lies in the investment of R&D personnel. In the generative large model AI 2.0 era, the cost of model production mainly lies in the investment of computing resources. As the scaling law is verified in large language models, multimodal models, video generation models, and slow thinking and reasoning processes, the cost of producing and using large models can be directly equivalent to the consumption of computing resources. In short, the popularization and commercialization of generative large model AI must be committed to reducing the production and use costs of large models, which will inevitably require the iteration and optimization of computing power in combination with large models, and it is also necessary to iterate the design and application of large models according to the characteristics of computing resources.

Therefore, in the field of generative big model AI, SenseTime's core strategy is to achieve seamless integration of computing power large devices (SCO), big models and applications (CNI), so that applications drive models and models drive computing power optimization. We have established a trinity strategy of "big devices-big models-applications" to improve the efficiency of computing power resources and serve our customers well through order-of-magnitude optimization. We are committed to becoming the big model service provider that understands computing power the most, and the computing power service provider that understands big models the most.

In the field of traditional AI, we will make full use of our core capabilities in visual perception and multimodal models, concentrate resources, clarify direction, and serve both domestic and international markets through a set of R&D investments.

At the organizational level, we will build a more centralized and efficient organizational structure around strategy and core resources, promote the concentration and intensive investment of resources, and accelerate the process of rejuvenation of organization and management. As SenseTime celebrates its 10th anniversary and the era of general artificial intelligence is approaching, let us regain our original entrepreneurial spirit.

I would like to thank every SenseTime employee for their dedication and hard work over the past decade. I hope we can continue to maintain our past tenacity, courage, and optimism, actively embrace change, and unswervingly explore the path of "technological originality and industrial value closed loop". In the next decade, we will work hard to jointly define and create the era of general artificial intelligence!

Xu Li

10/22/2024