1. [Iflytek is turning from G end to C end and B end]
On December 18th, Iflytek developed a large-scale model on the platform of national productivity, and cooperated with Huawei to launch "Feixing No.1" and "Feixing No.2", aiming to surpass OpenAI. The company is facing financial pressure. In 2023, the R&D investment reached 3.84 billion yuan, with a loss of net profit and negative operating cash flow. Iflytek is shifting from G-end to C-end and B-end, and the business of learning machines and smart cars at C-end has grown significantly. The AI industry is fiercely competitive, investors are cautious and the price war is intensifying. Relying on the advantages of full-stack self-research and independent control, Iflytek won the favor of big financial institutions and state-owned enterprises and became the representative of domestic substitution.
2.[Volcano Engine Video Cloud and Lexin Technology and ToyCity jointly launched the AI+ Hardware Smart Jump Program]
On December 18th, Volcano Engine Video Cloud, Lexin and ToyCity jointly launched the AI+ Hardware Smart Leap Program, which combined the advantages of bean bag model, Volcano Engine’s anthropomorphic voice dialogue, ToyCity’s trendy play design, and the AI chip of Lexin Technology to promote the popularization of AI trendy play.
3.[The "Tongyi" application team will be split from Alibaba Cloud, and insiders: Alibaba Cloud still retains Tongyi B-side business]
On December 18th, the "Tongyi" application team will be split from Alibaba Cloud and merged into Ali Intelligent Information Business Group. According to people familiar with the matter, the client and team of Tongyi to C will be merged into the Alibaba Group, while Tongyi to B and the open source part are still in the system of Alibaba Cloud. According to previous media reports, after the adjustment, Tongyi PC and App teams are on the same level as the intelligent search product "Quark", and the original Tongyi laboratory remains in the Alibaba Cloud system. This move can be seen as a step for Ali to sort out the internal AI to C application.
1. [First appearance of 3D generation model of bean curd: physical world simulation simulator supporting AIGC creation]
On December 18th, at the Force conference in Volcano Engine, the 3D generation model of bean curd also made its debut. Combined with veOmniverse, a digital twin platform in Volcano Engine, this model can efficiently complete intelligent training, data synthesis and digital asset production, and become a set of physical world simulation simulator supporting AIGC creation.
2.[Meta launches Apollo open source model to make AI "understand" videos]
On December 18th, Meta cooperated with Stanford University to launch the Apollo series AI model to improve the machine’s understanding of video. Apollo model consists of two components, which process video frames and track the changes of objects and scenes respectively, and add timestamps between video clips to keep time perception. The model adopts phased training method and emphasizes the importance of training method. Meta optimizes data combination, and finds that the balance between 10-14% text data and video content can improve language understanding and video processing ability.
1. ["AI behemoth" Databricks won $10 billion in financing, and its valuation rose to $62 billion]
On December 18th, the AI unicorn company Databricks successfully raised $10 billion with a valuation of $62 billion. This financing has become one of the largest rounds of venture capital in history. Databricks is expected to achieve positive free cash flow by the end of January next year, and plans to break through the annualized income of $3 billion in January, and the income in the following fiscal year is expected to be $3.8 billion. The company plans to use funds to attract top AI talents, invest in new AI products, and provide funds for mergers and acquisitions of potential startups. This financing provides Databricks with the flexibility to postpone the initial public offering (IPO). Headquartered in San Francisco, Databricks provides data analysis services to 10,000 customers, and currently has 7,000 employees, and is competing with Snowflake.
2.[Investment and financing in the field of artificial intelligence are hot. The amount of financing in the year has exceeded 82 billion yuan]
On December 18 th, this year, the wave of AI (artificial intelligence) swept the world. China’s investment and financing activities in the AI field continue to be hot, driving a large amount of capital into the industry. The data shows that as of December 17, there were 644 investment and financing incidents in the AI field during the year, exceeding the whole year of 2023 (633); The amount involved was 82.129 billion yuan, compared with 63.676 billion yuan in 2023.
3.[Aouita successfully completed the C round of financing, raising more than 11 billion yuan]
On December 18th, Aouita Science and Technology announced the completion of the C round of financing, raising more than 11 billion yuan. This round of financing is jointly funded by a number of investment institutions. Aouita actively promoted capital cooperation with its partners this year, including investing in Huawei’s subsidiary "Yinwang" and signing a strategic cooperation agreement. The two sides will cooperate deeply in the field of software-defined vehicles. Aouita has released three strategic products and plans to launch the fourth one next year. The company’s sales volume exceeded 10,000 for two consecutive months, and made progress in channel layout, overseas market expansion and technology research and development. Aouita plans to complete the investment delivery of Huawei in the first half of next year, and prepare to go public in IPO in 2026.
1. [YouTube technology will be able to identify AI replicas of celebrities and creators]
On December 18 th, YouTube cooperated with Creative Artists Agency (CAA) to help creators identify and delete the content that uses AI to generate portraits on the platform. The company plans to test these controls with celebrities and athletes early next year, and then launch them to top creators and professionals. YouTube has previously announced that it will launch tools to manage AI depictions of creators and their voices, and allow celebrities and creators to manage AI copies of their portraits on a large scale. CAA has launched CAAVault to scan and store digital portraits of customers. YouTube is also developing synthetic singing recognition technology to detect AI content that replicates the singer’s voice and allow music companies to request the deletion of AI content that simulates the artist’s voice.
2[Google: As long as there is manual supervision, customers can use their AI to make decisions in "high risk" areas]
On December 18 th, Google updated its usage policy, allowing the use of generative artificial intelligence tools for automatic decision-making in high-risk areas under manual supervision. The new policy allows automatic decisions that may affect individual rights in areas such as employment, housing, insurance and social welfare, as long as there is human supervision. Google’s previous draft articles once said that high-risk automated decision-making should be completely banned, but in fact it has never been banned, only requiring manual supervision. A Google spokesperson said that manual supervision requirements have always existed and the terms have been reclassified to make it clearer to users. Regulators are concerned about the potential bias of AI system, and research shows that AI may aggravate the problem of historical discrimination.
