Summary of Conference Call on Data Processing in Autonomous Driving Industry Overview - The discussion focused on the application of data processing in the autonomous driving sector, highlighting the increasing importance of data handling technologies due to the exponential growth of data volume as autonomous driving technology advances [2][9]. Key Points and Arguments - **Data Processing Demand**: The need for data cleaning, labeling, and preprocessing is rising, necessitating advanced algorithms and technology platforms to enhance processing efficiency and reduce operational costs [2][9]. - **Technical Barriers**: There are significant technical barriers in achieving high-precision data labeling and processing, emphasizing the importance of data quality [2][9]. - **Collaboration with Automakers**: Collaborating with automotive manufacturers is seen as an effective way to address data standardization issues, which is crucial for advancing autonomous driving technology [2][4]. - **Revenue Growth Outlook**: The company anticipates revenue and profit growth in the coming years, reflecting an optimistic market outlook despite previous challenges [2][24]. Data Processing Techniques - **Algorithm Optimization**: The use of clustering algorithms and deep learning techniques is being employed to optimize object detection frameworks, enhancing tracking efficiency and accuracy [3][10]. - **AI Data Services Focus**: The company is expanding its revenue from autonomous driving-related services, aiming to maintain a 15% to 20% share of total revenue from this segment [4]. Industry Trends - **End-to-End Algorithm Development**: The shift towards end-to-end algorithms in autonomous driving is significantly impacting the data services industry, leading to explosive data demand and increased labeling complexity [5][10]. - **Data Accumulation**: Data accumulation is primarily through self-collection and collaboration with clients, with a strong emphasis on data quality and ownership protection [5][10]. Business Model and Revenue Composition - **Service Models**: The company offers two main service models: project-based services tailored to client needs and product-based services through the sale of mature data products, which together account for over 90% of annual revenue [6][16]. - **Revenue Structure**: The revenue is mainly derived from one-time project income and periodic income, with project income constituting about 90% of annual revenue [16]. Data Quality and Compliance - **Real vs. Synthetic Data**: Real data is deemed significantly more valuable than synthetic data for AI model training, although synthetic data can supplement areas where real data is lacking [7][17]. - **Data Compliance Challenges**: Data compliance issues are a major challenge for the application of synthetic data, necessitating exploration of data simulation techniques to address compliance while unlocking new business opportunities [18]. Future Trends and Challenges - **Data Demand Drivers**: Factors driving the increasing data demand in autonomous driving include vehicle diversity, sensor types, and advancements in driving technology [21]. - **Competitive Advantage**: The company possesses competitive advantages in data processing, including platform capabilities, algorithm efficiency, project management, and resource allocation [22]. Market Dynamics - **Collaboration vs. Competition**: As automakers enhance their data collection capabilities, the relationship between the company and automakers is viewed as collaborative rather than competitive, focusing on data processing and compliance [20]. - **Government Initiatives**: Recent government initiatives aim to open up public data, which could facilitate data utilization in the industry [19]. Financial Outlook - **Recovery from Previous Decline**: The company expects to recover from a rare financial decline experienced last year, with Q1 data showing significant revenue and profit rebounds, aiming for stable growth throughout the year [24].

Group 1: Financial Performance - In 2023, the company's revenue was 170 million yuan, a decrease of 35.33% compared to the previous year, primarily due to overseas clients undergoing layoffs and adjustments in business direction and R&D cycles [3] - The domestic business faced challenges despite macroeconomic stabilization, with clients showing caution in centralized R&D investments, leading to a slowdown in budget releases and increased competition in the industry [3][4] Group 2: Revenue Growth Factors - The revenue growth in Q1 2024 was driven by the gradual development and application of large model technology, resulting in increased demand for data in AI research and development [4] - The company is currently engaging with regions such as Beijing, Anhui, Shandong, and Hebei for potential projects, with disclosures to follow if cooperation meets requirements [4][5] Group 3: Data Demand in Large Models - The data demand during the pre-training phase of large models differs from traditional training data in terms of scale, quality, and sources, with pre-training data typically in the tens of billions of tokens compared to around 1 billion for traditional models [5] - Future data demand is expected to continue growing significantly as more large model products are launched and iterated, particularly for high-quality, large-scale copyright data and public data [6][7] Group 4: Trends in Reinforcement Learning - The overall data demand for reinforcement learning is gradually increasing, with a trend towards expanding into more verticals such as law, finance, and healthcare [7] - There is a shift from single-modal to multi-modal data requirements, with a focus on text-video and text-image combinations in annotation tasks [8]

北京市天元律师事务所 关于北京海天瑞声科技股份有限公司 2023 年年度股东大会的法律意见 京天股字（2024）第 369 号 致：北京海天瑞声科技股份有限公司 北京海天瑞声科技股份有限公司（以下简称"公司"）2023 年年度股东大会 （以下简称"本次股东大会"）采取现场投票与网络投票相结合的方式，现场会议 于 2024 年 6 月 7 日 13:30 在北京市海淀区知春路 68 号院 1 号楼 4 层 401 召开。 北京市天元律师事务所（以下简称"本所"）接受公司聘任，指派本所律师参加本 次股东大会现场会议，并根据《中华人民共和国公司法》《中华人民共和国证券 法》（以下简称"《证券法》"）、《上市公司股东大会规则》（以下简称"《股东大会 规则》"）以及《北京海天瑞声科技股份有限公司章程》（以下简称"《公司章程》"） 等有关规定，就本次股东大会的召集、召开程序、出席现场会议人员的资格、召 集人资格、会议表决程序及表决结果等事项出具本法律意见。 为出具本法律意见，本所律师审查了《北京海天瑞声科技股份有限公司第二 届董事会第二十四次会议决议公告》《北京海天瑞声科技股份有限公司第二届监 事会第二十三次会议决议公告 ...

证券代码：688787 证券简称：海天瑞声 公告编号：2024-029 重要内容提示： 本次会议是否有被否决议案：无 北京海天瑞声科技股份有限公司 2023 年年度股东大会决议公告 本公司董事会及全体董事保证公告内容不存在任何虚假记载、误导性陈述 或者重大遗漏，并对其内容的真实性、准确性和完整性依法承担法律责任。 | 1、出席会议的股东和代理人人数 | 10 | | --- | --- | | 普通股股东人数 | 10 | | 2、出席会议的股东所持有的表决权数量 | 23,936,483 | | 普通股股东所持有表决权数量 | 23,936,483 | | 3、出席会议的股东所持有表决权数量占公司表决权数量的 | 39.9706 | | 比例（%） | | | 普通股股东所持有表决权数量占公司表决权数量的比例 | 39.9706 | | （%） | | （四） 表决方式是否符合《公司法》及公司章程的规定，大会主持情况等。 一、 会议召开和出席情况 （三） 出席会议的普通股股东、特别表决权股东、恢复表决权的优先股股东及 其持有表决权数量的情况： （一） 股东大会召开的时间：2024 年 6 月 7 日 （二） ...

Group 1: Data Service Trends - The demand for data services in reinforcement learning is increasing, with a trend towards more verticals such as law, finance, and healthcare [3] - Evaluation metrics for reinforcement learning annotations are becoming more diverse, requiring annotators to assess models from multiple dimensions [3] - The shift from unimodal to multimodal data annotation is evident, with a focus on text-image and text-video combinations [3][4] Group 2: Automation in Data Annotation - Current data annotation tasks in large models primarily focus on supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), with a strong reliance on human input [4] - Some projects have begun to implement algorithmic pre-annotation strategies to enhance the efficiency of manual annotation and verification [4] Group 3: Multimodal Development and Data Needs - The transition to multimodal models will create new data demands, such as generating images from text inputs, requiring machines to understand and map keywords to image tags [5] - The importance of high-quality multimodal training datasets will increase, driving growth in the data service industry [5] Group 4: Synthetic Data Considerations - Synthetic data is viewed as a necessary byproduct of AI development, serving as an effective supplement for data collection, though it has limitations in replicating real-world features [5][6] - Most companies still rely on real-world data for model training, but they will monitor advancements in synthetic data technology to adjust their business strategies accordingly [6] Group 5: Copyright Data and Value Proposition - The value of the company lies in aggregating diverse copyright data, cleaning it, and providing tailored services based on client needs [6] - High-quality cleaning of copyright data is essential before it can be used for model training, ensuring compliance with legal standards [6] Group 6: Differences in Data Requirements - The data requirements for pre-training large models are similar to traditional deep learning but differ in scale, quality, and sources [7][8] - Pre-training data typically involves token counts in the hundreds of billions, compared to around 1 billion for traditional models, necessitating a richer variety of data sources [7][8]

证券代码：688787 证券简称：海天瑞声 北京海天瑞声科技股份有限公司 投资者关系活动记录表 编号：2024-008 投资者关系活动类别 □特定对象调研 □分析师会议 □媒体采访 √业绩说明会 □新闻发布会 □路演活动 □现场参观 □电话会议 □其他 （请文字说明其他活动内容） 参与单位名称及人员姓 参加业绩说明会的广大投资者 名 ...

Financial Performance - 2023 revenue was 1.70 billion yuan, a decrease of 35.33% year-over-year [3] - Overseas revenue declined significantly due to customer layoffs, business adjustments, and data export regulations [3] - Domestic revenue also decreased due to cautious R&D spending by customers and intensified competition [3] R&D Investment - R&D investment decreased in 2023 as the company's strategic focus on intelligent data production and standardized dataset products stabilized [3] - Investment in intelligent driving data platforms like DOTS-AD and integrated data processing platforms has matured, leading to reduced R&D intensity [4] - Standardized dataset product construction peaked in the first two years post-IPO and concluded in May 2023, resulting in a natural decline in investment [4] Automation and AI Impact - Automation in data labeling is a trend, but it aims to enhance human-machine collaboration rather than replace human effort [4] - The company has increased R&D efforts to improve intelligent data production capabilities [4] - Multimodal development in AI models will create new data demands, expanding the data service industry's growth potential [5] Big Model Data Progress - The company is developing datasets for large language models, voice models, and visual models [6] - Collaborations with major domestic big model manufacturers cover RLHF data scoring, prompt rewriting, and human-machine interaction [6] - Efforts are underway to expand data service capabilities, including data cleaning, model evaluation, and fine-tuning [6]

华泰联合证券有限责任公司 关于北京海天瑞声科技股份有限公司 2023 年度持续督导跟踪报告 1 长。以上因素共同导致 2023 年度归属于母公司所有者的净利润、归属于母公司 所有者的扣除非经常性损益的净利润显著下滑并且出现亏损。上述不利因素目前 已有所改善，但如果公司收入增长无法覆盖各类投入及期间费用支出，公司业绩 存在下滑或亏损的风险。 （二）核心竞争力风险 | 保荐机构名称：华泰联合证券有限责任公司 | 被保荐公司简称：海天瑞声 | | --- | --- | | 保荐代表人姓名：张鹏 | 联系电话：010-56839300 | | 保荐代表人姓名：葛青 | 联系电话：010-56839490 | 根据《证券法》《证券发行上市保荐业务管理办法》《上海证券交易所科创 板股票上市规则》等有关法律、法规的规定，华泰联合证券有限责任公司（以下 简称"华泰联合证券"或"保荐机构"）作为北京海天瑞声科技股份有限公司（以下 简称"海天瑞声"、"公司"或"发行人"）首次公开发行股票的保荐机构，对海天瑞 声进行持续督导，并出具本持续督导跟踪报告。 一、保荐机构和保荐代表人发现的问题及整改情况 无。 二、重大风险事项 公司目 ...

Group 1: Company Performance and Revenue Growth - In Q1 2024, the company's revenue growth is attributed to the increasing investment in AI technology by global tech companies, leading to a significant rise in data demand, particularly in multilingual intelligent voice and text services [3] - The company has accumulated over 1,550 proprietary standardized training data products by the end of December 2023, positioning it among the top global enterprises in terms of database inventory [5] Group 2: Core Competencies and Competitive Advantages - The company's core competitive advantages include: - **Technical Platform Capability**: Increased R&D investment has enhanced algorithm, platform, and engineering capabilities, improving human-machine collaboration efficiency [4] - **Business Model**: The dual service-product model significantly contributes to revenue and gross profit, ensuring scalability and high profit margins [5] - **Supply Chain Resource Management**: Long-term development of a supply chain system ensures resource acquisition and supports customer expansion [5] - **Data Security and Compliance**: Established a mature security and compliance management system, achieving certifications such as ISO/IEC 27001 and compliance with GDPR and local laws [6] Group 3: Future Trends and Technological Impact - Automation in data labeling is seen as a trend that enhances human-machine collaboration rather than replacing human roles, with ongoing investments in intelligent data production [6] - Synthetic data is viewed as a necessary evolution in AI, serving as a supplementary method for data collection, but it cannot replace real-world data for model training [7] - The shift towards multimodal large models is expected to create new data demands, emphasizing the importance of high-quality multimodal training datasets [8] Group 4: Integration of Large Models - The company is exploring the integration of large models into its data processing platform to enhance efficiency in data handling and production [9] - Initial research and planning have been conducted on pre-training datasets for large models, with efforts to acquire and clean relevant data [9]

关于收到上海证券交易所《关于终止对北京海天瑞声科技 股份有限公司向特定对象发行股票审核的决定》的公告 本公司董事会及全体董事保证本公告内容不存在任何虚假记载、误导性陈述 或者重大遗漏，并对其内容的真实性、准确性和完整性依法承担法律责任。 北京海天瑞声科技股份有限公司（以下简称"公司"）于 2024 年 4 月 25 日召开公司第二届董事会第二十四次会议、第二届监事会第二十三次会议，审 议通过了《关于终止 2023 年度向特定对象发行 A 股股票事项的议案》，同意公 司终止本次向特定对象发行 A 股股票事项，并向上海证券交易所申请撤回本次 发行的申请文件。具体内容详见公司于 2024 年 4 月 26 日在上海证券交易所 （以下简称"上交所"）网站（www.sse.com.cn）及指定的信息披露媒体披露 的相关公告。 2024 年 4 月 29 日，公司与保荐机构华泰联合证券有限责任公司向上交所 提交了《北京海天瑞声科技股份有限公司关于撤回 2023 年度向特定对象发行 A 股股票并在科创板上市申请文件的申请》和《关于撤销北京海天瑞声科技股份 有限公司 2023 年度向特定对象发行 A 股股票并在科创板上市保荐 ...