This appears to be a transcript of a presentation on AI trends, specifically focusing on agentic workflows, visual AI, and large language models. The speaker discusses several key points, including:

  1. Agentic Workflows: The concept of building AI systems that can perform tasks that are typically done by humans, such as reading text or images and generating text.
  2. Visual AI: The ability to process and analyze visual data from images and videos, which is becoming increasingly important for various applications.
  3. Large Language Models: The advancements in language models that enable them to not only answer human questions but also perform tasks like tool use.

The speaker highlights four key trends:

  1. Agentic AI: The ability of agentic workflows to speed up token generation and perform tasks more efficiently.
  2. Tuning Large Language Models: The development of large language models that can support tool use, enabling agentic workflows to perform a wider range of tasks.
  3. Data Engineering’s Importance: The increasing need for data engineering expertise in managing unstructured data and deploying it effectively.
  4. The Text Processing Revolution: The text processing revolution has already arrived, while the image processing revolution is just beginning, which will enable many businesses to get more value out of visual data.

The speaker concludes by emphasizing that this is a great time to be a builder, with agentic AI expanding the set of possible applications and enabling new uses for visual AI.

Translation

AI 趨势的演讲内容,重点讨论了以下几点: 1. **Agentic Workflows**:构建能执行人类任务的AI系统,如文本或图像阅读和生成文本。 2. **Visual AI**:处理和分析图像和视频中的视觉数据的能力,这对于多个应用程序变得越来越重要。 3. **Large Language Models**:语言模型的进展,使其不仅可以回答人类问题,而且还能执行工具使用。 演讲者提到以下四个主要趋势: 1. **Agentic AI**:agentic 流程加速令牌生成并实现更高效的任务。 2. **Large Language Models 的调优**:支持工具使用的大型语言模型的发展,使agentic 流程能够完成更多任务。 3. **数据工程师的重要性**:管理和部署未结构化数据的需求日益增加。 4. **文本处理革命已经开始,而图像处理革命才刚刚开始**:这使得许多企业可以更好地利用视觉数据。 演讲者总结指出,这是成为一个构建师的理想时期,agentic AI 扩大了可能的应用范围,使新工具使用和视觉 AI 的新用途成为可能。 #### Reference: https://www.youtube.com/watch?v=KrRD7r7y7NY
<
Previous Post
Can Test-Time-Training (TTT) fix the Scaling law ceiling
>
Next Post
Richard Sutton: reinforcement learning and continuous training