Author: XiaoFei

Anthropic Launches Transparency Hub Anthropic has unveiled its [Transparency Hub](https://www.anthropic.com/transparency), a comprehensive resource detailing the company’s efforts to ensure its AI systems are safe, beneficial, and trustworthy. The hub offers insights into Anthropic’s methodologies for model evaluation, safety testing, platform abuse detection, internal governance, societal impact assessment, research initiatives, and security/privacy safeguards. This initiative aims to provide clear insights into responsible scaling, safety protocols, and risk mitigation, especially with evolving regulatory frameworks. The hub includes reports on banned accounts, appeals, government requests, and other key metrics. Anthropic plans to continuously update the hub, reflecting best practices and addressing emerging challenges as AI technology advances. The goal is to provide a structured and accessible view of Anthropic’s model development and deployment for users, policymakers, and stakeholders, with feedback welcome at transparency@anthropic.com. https://www.anthropic.com/news/introducing-anthropic-transparency-hub

Anthropic 推出透明度中心 Anthropic 推出了其 [透明度中心](https://www.anthropic.com/transparency)，这是一个全面的资源，详细介绍了该公司为确保其人工智能系统安全、有益且值得信赖所做的努力。该中心提供了对 Anthropic 模型评估、安全测试、平台滥用检测、内部治理、社会影响评估、研究计划以及安全/隐私保护方法论的见解。 该举措旨在提供对负责任的规模化、安全协议和风险缓解的清晰见解，尤其是在不断发展的监管框架下。该中心包括关于被禁账户、申诉、政府请求和其他关键指标的报告。Anthropic 计划持续更新该中心，反映最佳实践并应对随着人工智能技术进步而出现的新挑战。 其目标是为用户、政策制定者和利益相关者提供对 Anthropic 模型开发和部署的结构化和可访问的视角，欢迎通过 transparency@anthropic.com 提供反馈。 https://www.anthropic.com/news/introducing-anthropic-transparency-hub

Google is releasing Gemini 2.0 Flash-Lite, now generally available in Gemini API, promising improved performance over previous models, especially in reasoning, multimodal tasks, and factuality. This update also features simplified pricing, making it cost-effective for projects utilizing long context windows. Developers are already leveraging the speed and efficiency of Gemini 2.0 Flash, with examples in voice AI by Daily.co, data analytics by Dawn, and video editing by Mosaic. The new model is accessible in Google AI Studio and Vertex AI, with a focus on enhancing application development across various sectors. https://deepmind.google/discover/blog/start-building-with-gemini-20-flash-and-flash-lite/

谷歌正在发布 Gemini 2.0 Flash-Lite，现已在 Gemini API 中全面推出，承诺在推理、多模态任务和事实性方面比之前的模型有所改进。此次更新还简化了定价，使其对于使用长上下文窗口的项目具有成本效益。开发人员已经在利用 Gemini 2.0 Flash 的速度和效率，Daily.co 在语音 AI 方面，Dawn 在数据分析方面，Mosaic 在视频编辑方面都提供了实例。新模型可在 Google AI Studio 和 Vertex AI 中访问，重点是加强跨各行业的应用程序开发。 https://deepmind.google/discover/blog/start-building-with-gemini-20-flash-and-flash-lite/

Google’s developer blog announced the general availability of Gemini 2.0 Flash-Lite in the Gemini API. This model offers improved performance over 1.5 Flash across various benchmarks, with simplified pricing for prompts exceeding 128K tokens. The report highlighted applications like Daily, which uses Gemini 2.0 Flash-Lite for voice AI, improving speed and accuracy in detecting voicemail. Additionally, Dawn is leveraging Gemini 2.0 Flash for data analytics, significantly reducing search times and costs. Mosaic is utilizing Gemini 2.0 Flash for video editing, accelerating tasks with its long-context capabilities and cost-effective pricing. The blog encourages developers to utilize the Gemini 2.0 Flash family for building innovative applications. https://deepmind.google/discover/blog/start-building-with-gemini-20-flash-and-flash-lite/

谷歌开发者博客宣布 Gemini API 中 Gemini 2.0 Flash-Lite 现已全面上市。与 1.5 Flash 相比，该模型在各种基准测试中表现更出色，并且简化了超过 128K 标记的提示的定价。报告重点介绍了 Daily 等应用，该应用使用 Gemini 2.0 Flash-Lite 进行语音 AI，从而提高了语音邮件检测的速度和准确性。此外，Dawn 正在利用 Gemini 2.0 Flash 进行数据分析，从而显著缩短了搜索时间并降低了成本。Mosaic 正在使用 Gemini 2.0 Flash 进行视频编辑，凭借其长上下文能力和具有成本效益的定价，加速了任务处理。该博客鼓励开发人员利用 Gemini 2.0 Flash 系列构建创新应用。 https://deepmind.google/discover/blog/start-building-with-gemini-20-flash-and-flash-lite/

Gemini 2.0 Model Updates Gemini 2.0 is now available to everyone Google DeepMind has announced significant updates to the Gemini 2.0 model family. These updates include the general availability of the updated Gemini 2.0 Flash via the Gemini API in Google AI Studio and Vertex AI, allowing developers to build production applications. The company is also releasing an experimental version of Gemini 2.0 Pro, optimized for coding and complex prompts, and a new model, Gemini 2.0 Flash-Lite, designed for cost efficiency. All models will initially feature multimodal input with text output, with more modalities planned. The release also highlights ongoing investments in safety measures for the Gemini model family. https://deepmind.google/discover/blog/gemini-2-0-is-now-available-to-everyone/

Gemini 2.0 模型更新 Gemini 2.0 现已面向所有人开放 Google DeepMind 宣布了 Gemini 2.0 模型家族的重大更新。这些更新包括通过 Google AI Studio 和 Vertex AI 中的 Gemini API 普遍提供更新后的 Gemini 2.0 Flash，使开发人员能够构建生产应用程序。 该公司还将发布 Gemini 2.0 Pro 的实验版本，该版本针对编码和复杂提示进行了优化，以及一个新模型 Gemini 2.0 Flash-Lite，该模型旨在实现成本效益。 所有模型最初都将具有多模态输入和文本输出，并计划支持更多模态。此次发布还强调了对 Gemini 模型家族安全措施的持续投资。 https://deepmind.google/discover/blog/gemini-2-0-is-now-available-to-everyone/

### Frontier Safety Framework 2.0 DeepMind has released an updated Frontier Safety Framework (FSF) to address potential risks associated with advanced AI models, particularly as they approach AGI. The framework emphasizes stronger security protocols, especially for models with critical capabilities, recommending heightened security levels to prevent unauthorized access and exfiltration of model weights. Key updates include enhanced procedures for deployment mitigations, ensuring thorough safety assessments and corporate governance reviews. The framework also outlines an industry-leading approach to addressing the risks of deceptive alignment, including automated monitoring and proactive research into mitigation strategies. The ultimate goal is to establish common standards and best practices for evaluating and securing the benefits of future AI models, with a collaborative approach involving researchers, companies, and governments. https://deepmind.google/discover/blog/updating-the-frontier-safety-framework/

### 前沿安全框架 2.0 DeepMind发布了更新后的前沿安全框架（FSF），以应对与先进人工智能模型相关的潜在风险，特别是在这些模型接近通用人工智能（AGI）之际。该框架强调更强大的安全协议，特别是针对具有关键能力的模型，建议提高安全级别以防止未经授权的访问和模型权重的泄露。 关键更新包括增强的部署缓解程序，确保彻底的安全评估和公司治理审查。该框架还概述了一种行业领先的方法来解决欺骗性对齐的风险，包括自动化监控和主动研究缓解策略。最终目标是为评估和保障未来人工智能模型的益处建立通用标准和最佳实践，并采取由研究人员、公司和政府共同参与的合作方式。 https://deepmind.google/discover/blog/updating-the-frontier-safety-framework/