当前，税务领域的领导者正面临前所未有的挑战，包括地缘政治的不确定性、不断演变且逐渐碎片化的法规体系、税务人才的短缺以及技术的迅速变革 等。在这一背景下，税务管理模式正面临着来自内部和外部的双重压力，亟需进行大规模投入与转型，来应对未来的复杂局势。为此，本报告重点强调了 三项战略要务，并详述所需的关键步骤，助力企业在未来竞争中脱颖而出。（下） 应用GenAI（生成式人工智能）为税务工作服务 正如毕马威国际的税务与法律人工智能专家所指出的，"在一个不断变化且资源有限的世界里，企业的税务部门正面临着巨大的压力，而生成式人工智能 显然具备大幅提升效率和生产力的强大能力。" 人工智能可以简化税务合规和报告流程，减少这些任务占用的时间和精力。它还可以增强对信息（如合同内容）的分析，促进在税收争议时与税务机关的 有效沟通。通过扫描和总结大量数据，生成式人工智能可以提供对全球税收环境的重要洞察，使组织能够了解最新情况并及时响应变化。 此外，生成式人工智能通过识别业务模式和变化趋势，帮助处理复杂的税务交易，从而发现潜在的风险（如隐性税务负债）和机会。这一技术进步不仅提 高了工作效率，还使得税务部门能够专注于战略决策和增值活动。 ...

Core Viewpoint - The Shanghai Cyberspace Administration has initiated penalties against several generative artificial intelligence service websites for failing to comply with legal safety assessment requirements and for not implementing necessary safety measures to prevent the generation of illegal content [1] Group 1: Regulatory Actions - The Shanghai Cyberspace Administration discovered that certain websites providing generative AI services did not conduct safety assessments as mandated by law [1] - These websites failed to take necessary precautions to prevent the generation of illegal content, including violations of personal information rights and the production of illicit materials such as money laundering content and pornographic images [1] - Companies are urged to take down related functionalities and may only resume operations after passing safety evaluations [1] Group 2: Future Enforcement Focus - The Shanghai Cyberspace Administration will continue to combat the misuse of AI, particularly focusing on issues related to "AI disguise," "AI face-swapping and voice-changing," and "AI forgery" [1] - There will be a concentrated effort to address algorithmic services that infringe on personal information rights, with strict penalties for repeat offenders and those with serious issues [1]

Core Viewpoint - Generative AI is rapidly integrating into the digital lives of minors, presenting both opportunities and significant risks, particularly concerning the generation of inappropriate content and the potential for data pollution in AI models [1][3][5] Group 1: AI Applications and Risks - AI image generation applications can easily produce sensitive images involving minors, raising serious safety concerns [1][3] - Some applications in the market are operating in a "gray market," lacking proper regulation and oversight, which increases the risk of misuse [5][8] - A recent investigation revealed that certain AI applications can generate explicit images of minors without restrictions, highlighting the urgent need for regulatory measures [3][5] Group 2: Regulatory Actions and Compliance - The Central Cyberspace Affairs Commission is actively addressing issues related to the misuse of AI technology, particularly concerning the creation of pornographic content and the protection of minors [2] - There are currently over 300 AI image generation applications available, with varying levels of content quality and compliance [5] - Some applications have implemented measures to intercept sensitive content, but risks remain due to the potential for data pollution in open-source models [7][8] Group 3: Expert Insights and Recommendations - Experts emphasize the importance of incorporating "minor protection" as a core value in the development of AI applications [1][5] - Recommendations include pre-cleaning data, adversarial training, and establishing compliance guidelines to mitigate risks associated with AI-generated content [7][8] - The introduction of regulations, such as the "Identification Method for AI-Generated Synthetic Content," aims to enhance accountability and traceability of non-compliant content [7][8]

Core Viewpoint - The article discusses the development of a new framework, RFpeptides, for the de novo design of high-affinity macrocyclic peptides targeting proteins, utilizing advancements in deep learning and artificial intelligence [2][3][10]. Group 1: Background and Challenges - Traditional methods for peptide drug development rely on natural product discovery or high-throughput screening of random peptides, which are resource-intensive and limited in scope [6][8]. - The challenges in natural product discovery include difficulties in synthesis, poor stability, and low tolerance to mutations [6]. - High-throughput screening methods, while powerful, are time-consuming and costly, covering only a small fraction of the chemical diversity available in macrocyclic compounds [6][9]. Group 2: Innovations in Design Methodology - The RFpeptides framework allows for precise de novo design of macrocyclic peptides with high affinity for target proteins, addressing the limitations of previous methods [3][12]. - The research team expanded existing structural prediction networks and protein backbone generation frameworks to incorporate cyclic relative position encoding, enhancing the design process [12]. Group 3: Experimental Results - The team tested up to 20 designed macrocyclic peptides against four different proteins (MCL1, MDM2, GABARAP, and RbtA), achieving medium to high affinity binders for all targets [13]. - Notably, a high-affinity binder for RbtA was designed with a dissociation constant (K_d) of less than 10 nM based solely on predicted target structure [13]. - Structural analysis of the designed macrocyclic peptide complexes with MCL1, GABARAP, and RbtA showed high agreement with computational models, with Cα RMSD values less than 1.5 Å [14]. Group 4: Implications and Future Applications - The RFpeptides framework provides a systematic approach for the rapid custom design of macrocyclic peptides for diagnostic and therapeutic applications, indicating significant potential in the pharmaceutical industry [16].

Summary of the 2025 AI Index Report Industry Overview - The report focuses on the **artificial intelligence (AI)** industry, highlighting its rapid development and integration into various sectors, including healthcare and transportation [2][3][4]. Key Insights and Arguments 1. **Performance Improvements**: AI systems have shown significant performance improvements in benchmark tests, with scores for MMMU, GPQA, and SWE-bench increasing by **18.8%**, **48.9%**, and **67.3%** respectively from 2023 to 2024 [9][51]. 2. **Integration into Daily Life**: AI is increasingly integrated into everyday life, with **223 AI medical devices** approved by the FDA in 2023, a substantial increase from **6 in 2015**. Autonomous vehicles are also becoming more prevalent, with companies like Waymo providing over **150,000 rides** weekly [9][10]. 3. **Investment Surge**: In 2024, private investment in AI in the U.S. reached **$109.1 billion**, significantly higher than China's **$9.3 billion** and the UK's **$4.5 billion**. The growth in generative AI startups has led to an **18.7%** increase in investment [10]. 4. **Global AI Model Development**: The U.S. remains a leader in developing top AI models, with **40 models** created in 2024 compared to **15 in China** and **3 in Europe**. However, the quality gap is narrowing, with performance differences in key benchmarks decreasing significantly [10][53]. 5. **Responsible AI Practices**: There is a growing recognition of the need for responsible AI practices, but the adoption of standardized assessments remains low among major developers. Governments are taking more proactive steps to establish regulatory frameworks [11][12]. 6. **Public Sentiment**: Optimism about AI's benefits is rising globally, particularly in countries like China (83%) and Indonesia (80%), while skepticism persists in places like Canada (40%) and the U.S. (39%) [12]. 7. **Cost Efficiency**: The cost of using AI models has dramatically decreased, with the cost per million tokens for a GPT-3.5 level model dropping from **$20** to **$0.07** over 18 months, representing a reduction of over **280 times** [47]. 8. **Environmental Impact**: The carbon emissions from training AI models are increasing, with the training of models like GPT-4 emitting **5,184 tons** of CO2, compared to just **0.01 tons** for earlier models like AlexNet [50]. Other Important but Overlooked Content - **Educational Initiatives**: There is a notable increase in the implementation of computer science education globally, with two-thirds of countries now offering or planning to offer such education, although disparities in resource access remain [13]. - **AI Patent Growth**: The number of AI patents has surged from **3,833 in 2010** to **122,511 in 2023**, with China leading in total patents [49]. - **Hardware Advancements**: AI hardware is becoming faster, cheaper, and more energy-efficient, with performance improving at an annual rate of **43%** [49]. This comprehensive overview of the 2025 AI Index Report highlights the rapid advancements and challenges within the AI industry, emphasizing the need for responsible practices and the importance of public sentiment in shaping the future of AI.

Core Insights - Apple is in preliminary talks to acquire Perplexity, with executives like Adrian Perica and Eddy Cue involved, but the high valuation of $14 billion may deter Apple from proceeding with the acquisition [2][4] - Meta is also interested in Perplexity, having previously considered a deal before acquiring a stake in Scale AI for $14.8 billion [2][4] - Perplexity, founded in 2022, is recognized as the world's first conversational search engine, leveraging AI to provide direct, precise, and timely search results [5][6] Company Overview - Perplexity has gained significant attention from major tech companies like Apple and Meta, achieving a valuation of $14 billion within less than four years of its founding [4] - The company has been included in prestigious lists such as Fortune's Global AI Innovators and Forbes AI 50, highlighting its rapid growth and innovation in the AI search space [6] Market Trends - The rise of AI search engines like Perplexity is seen as a potential threat to traditional search engines, with a noted decline in Google search volume as users increasingly turn to AI services [8][9] - Perplexity has experienced rapid user growth, reaching 10 million monthly active users and processing 780 million searches by May 2023, with a projected growth to 1 billion queries per week within a year [9][11] Competitive Landscape - Despite its growth, Perplexity faces significant competition, particularly from Google, which has introduced AI features that mimic the capabilities of AI search engines [11] - Legal challenges regarding content scraping from major publishers pose additional risks to Perplexity's operational model [11] Conclusion - The rapid development of generative AI presents both opportunities and challenges for companies like Perplexity, attracting investment and interest from major players while navigating competitive and legal hurdles [12]

Core Insights - The article discusses the rise of generative artificial intelligence (AIGC) and its impact on various industries, particularly in film and digital content creation [1][12] - It highlights the emergence of new job roles such as generative AI directors and digital human trainers, driven by the rapid advancement of AI technologies [12][19] Group 1: Generative AI in Film Production - The article introduces Luo Chong, a generative AI director who transitioned from traditional film roles to utilizing AIGC for creative projects [1][3] - Luo's team employs AI tools for video creation, using features like "text-to-image" and "image-to-video" to produce short animations efficiently [5][7] - The team recently completed a complex project involving nearly 100 shots, showcasing the capabilities of AI in enhancing creative output [7] Group 2: New Job Roles and Industry Growth - The demand for AI-related job roles is increasing, with positions like AI digital human trainers becoming prominent in the workforce [12][19] - Companies are facing a talent shortage in AI applications, with most employees lacking formal backgrounds in the field, necessitating on-the-job learning [16][18] - The generative AI market in China is projected to reach hundreds of billions by 2024 and exceed one trillion by 2030, indicating significant growth potential [18] Group 3: Skills and Competition - The article emphasizes that while AI tools lower entry barriers for content creation, competition will shift towards value creation rather than mere tool proficiency [21] - Professionals in the industry are encouraged to develop creative thinking and innovation skills alongside technical proficiency in AI tools [21]

Core Viewpoint - The meeting highlighted the importance of technology, particularly ultra-high definition (UHD) and generative artificial intelligence, in transforming the broadcasting and online audio-visual industry, aiming to enhance user engagement and content quality [1][2]. Group 1: Technological Advancements - The broadcasting industry is experiencing significant changes due to the rise of UHD and generative AI, which provide new creative possibilities [1]. - The National Radio and Television Administration has set 2025 as the year for UHD development, with many TV series, web dramas, and documentaries expected to adopt UHD standards [1]. - The integration of technology and artistic creation is exemplified by the success of the 2025 Spring Festival Gala, which utilized advanced technology to enhance viewer engagement [1]. Group 2: Targeting Young Audiences - The rapid development of the internet and mobile internet has shifted young users' information consumption from large screens to smaller mobile devices, presenting a challenge for the broadcasting network [1]. - To attract young audiences, the industry is focusing on high-quality content, including UHD films, micro-short dramas, domestic and international sports events, and AR/VR games [2]. - A new television line has been established to synchronize the release of quality films and micro-short dramas, while also incorporating sports events and cultural documentaries [2]. Group 3: Comprehensive Service Offerings - The broadcasting sector is providing a comprehensive information service model, integrating "5G + TV + broadband + content + rights + X" to enhance user experience [2]. - Users can access classic programs and quality content through 5G services while on the go, and enjoy seamless integration with new set-top boxes at home [2]. - Smart broadcasting solutions are being developed to offer full-home intelligent access and services, catering to the needs of modern consumers [2].

Core Viewpoint - The article discusses the innovative UniGRF framework, which unifies retrieval and ranking tasks in recommendation systems using a single generative model, addressing inherent issues in traditional multi-stage recommendation paradigms [1][3][16]. Group 1: Pain Points of Traditional Recommendation Paradigms - Traditional recommendation systems typically employ a multi-stage approach, where a recall phase quickly filters a large item pool, followed by a ranking phase that scores and orders the candidates. This method, while efficient, often leads to information loss and performance bottlenecks due to the independent training of each phase [3][4]. - The separation of tasks can result in the premature filtering of potential interests outside the user's information bubble, causing cumulative biases and difficulties in inter-stage collaboration [3][4]. Group 2: Advantages of UniGRF - UniGRF integrates retrieval and ranking into a single generative model, allowing for full information sharing and reducing information loss between tasks [7]. - The framework is model-agnostic and can seamlessly integrate with various mainstream autoregressive generative model architectures, enhancing its flexibility [8]. - By maintaining a single model instead of two independent ones, UniGRF potentially improves efficiency in both training and inference processes [9]. Group 3: Key Mechanisms of UniGRF - The framework includes a Ranking-Driven Enhancer, which promotes effective collaboration between the recall and ranking phases by leveraging the high precision of the ranking outputs to guide the recall process [10][11]. - It also features a Gradient-Guided Adaptive Weighter that dynamically adjusts the weights of the loss functions for the two tasks based on their learning rates, ensuring synchronized optimization and overall performance enhancement [12]. Group 4: Experimental Results - Extensive experiments on three large public recommendation datasets (MovieLens-1M, MovieLens-20M, Amazon-Books) demonstrated that UniGRF significantly outperforms state-of-the-art (SOTA) models, highlighting the advantages of its unified framework [14][18]. - The framework shows particularly notable improvements in ranking performance, which is crucial as it directly impacts the quality of recommendations presented to users [18]. - Initial tests indicate that UniGRF adheres to the scaling law, suggesting potential performance gains with increased model parameters [18]. Group 5: Future Directions - The introduction of UniGRF offers a novel and efficient solution for generative recommendation systems, overcoming traditional multi-stage paradigm issues. Future research aims to expand the framework to include more recommendation stages and validate its large-scale applicability in real-world industrial scenarios [16][17].

OpenAI联合创始人兼首席执行官山姆·奥特曼在最新播客中披露，备受关注的GPT-5预计将于今年夏季发布，目前 具体发布日期尚未确定。随着GPT-5发布时间的临近，业界普遍认为，多模态大模型领域又将迎来新一轮的技术 竞争，该模型将成为生成式人工智能能力的一次重大升级。从早期测试者的反馈来看，其性能较GPT-4有显著提 升。但也有人担忧，从去年开始GPT-5就曾屡屡跳票，这会不会又是一次"狼来了"？ AI能力重大飞跃 OpenAI开启官方播客，CEO打头阵。当地时间6月18日，OpenAI发布了一则山姆·奥特曼的访谈视频。在40分钟的 专访中，奥特曼回应了大家普遍关心的GPT-5、隐私保护、广告业务、5000亿美元的投资项目"星际之门"等热点 话题。奥特曼说，GPT-5"可能是在今年夏天的某个时候"会发布，但他也同时表示，对于新模型，内部也在讨论 是简单地提升版本号，还是像GPT-4那样不断优化和改进。 奥特曼还暗示，GPT-5所代表的不仅仅是性能升级，它还可能标志着OpenAI朝着统一的、类似代理的模型迈出了 真正的第一步，此举将使其更接近其通用人工智能目标。"我认为我们已经接近这座山的尽头了"，他表示。 G ...