AI早报 2025年03月21日摘要：

• OpenAI语音模型更新：
  • OpenAI发布了三款新的语音模型：gpt-4o-transcribe、gpt-4o-mini-transcribe和gpt-4o-mini-tts。
• 腾讯混元模型进展：
  • 腾讯混元的推理模型T1将于今晚正式发布。
  • 混元模型已跻身Chatbot Arena排行榜全球Top15。
• 谷歌Chrome浏览器AI整合：
  • 谷歌Chrome浏览器将整合Gemini AI助手，以提升用户在线体验。
  • 用户可通过浏览器顶部图标启动该助手。
• Meta AI重返欧洲市场：
  • Meta AI重返欧洲市场，在WhatsApp等平台推出文本聊天功能。
  • 功能包括头脑风暴和旅行规划。
  • Meta承诺未使用欧盟用户数据进行模型训练。
• xAI图像生成API：
  • xAI推出图像生成API“grok-2-image-1212”。
  • 该API可根据描述生成最多10张JPG图像，每秒处理5次，单张收费0.07美元。
• Adobe PPT生成工具：
  • Adobe推出“Project Slide Wow”工具，利用生成式AI将客户数据转化为PPT。
  • 该工具可智能生成图表、标题和备注，并实时更新数据。
• OpenAI高性能推理模型：
  • OpenAI推出高性能推理模型O1-Pro API，提供视觉输入、函数调用等强大功能。
  • 该模型的调用费用高昂。

英文翻译：

AI Daily Briefing - March 21, 2025:

• OpenAI Voice Model Updates:
  • OpenAI has launched three new voice models: gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-mini-tts.
• Tencent Hunyuan Model Progress:
  • Tencent Hunyuan's inference model T1 will be officially released tonight.
  • The Hunyuan model has entered the global Top 15 on the Chatbot Arena leaderboard.
• Google Chrome AI Integration:
  • Google Chrome browser will integrate the Gemini AI assistant to enhance user online experience.
  • Users can launch the assistant through an icon at the top of the browser.
• Meta AI Returns to European Market:
  • Meta AI returns to the European market, launching text chat features on platforms like WhatsApp.
  • Features include brainstorming and travel planning.
  • Meta pledges not to have used EU user data for model training.
• xAI Image Generation API:
  • xAI launches the image generation API "grok-2-image-1212."
  • The API can generate up to 10 JPG images based on descriptions, processing 5 times per second, with a single image costing $0.07.
• Adobe PPT Generation Tool:
  • Adobe introduces the "Project Slide Wow" tool, using generative AI to convert customer data into PPTs.
  • The tool can intelligently generate charts, titles, and notes, and update data in real-time.
• OpenAI High-Performance Inference Model:
  • OpenAI launches the high-performance inference model O1-Pro API, providing powerful functions such as visual input and function calls.
  • The model's invocation costs are high.