加州大学伯克利分校推出「大世界模型」
支持百万 token 上下文，还能生成视频
加州大学伯克利分校近日推出名为「大世界模型」（Large World Model，简称LWM）的开源世界模型，在 GitHub 热榜榜首引起广泛关注。
LWM 的最大特点是其超大的上下文窗口长度，达到了 100 万 token，与谷歌同时推出的 Gemini 1.5 持平。这使其能够处理海量文本信息，在100 万 token 的文本中准确找到目标文本。
此外，LWM 还支持多模态信息处理，不仅能处理文本，还能处理图像、视频等非文本数据。它能够一口气看完 1 小时的视频，并从视频中提取关键信息。
LWM 的诞生标志着世界模型技术的发展进入了一个新的阶段。世界模型是一种人工智能模型，它通过对大量数据进行训练，构建了一个对世界的理解。LWM 的超大上下文窗口长度和多模态处理能力，使其对世界的理解更加全面和深入。
研究人员表示，LWM 可以应用于广泛的领域，包括自然语言处理、计算机视觉、视频理解等。它有望推动这些领域的突破性进展，并为人工智能技术带来新的可能性。
目前，LWM 已在 GitHub 上开源，供研究人员和开发者免费使用。这将进一步加速世界模型技术的研发和应用，为人工智能领域的创新注入新的活力。
英语如下：
**Headline: Berkeley Unveils Million-Token Contextual ‘Big World Model’**
**Keywords:** Large model, video generation, contextual understanding
**Article:**
The University of California, Berkeley has unveiled an open-source world modelcalled the “Large World Model” (LWM), which has garnered significant attention as the top trending repository on GitHub.
The key feature of LWM is its massive context window length of 1 million tokens, on par with Google’s recently released Gemini 1.5. This allows it to process vast amountsof text information, accurately locating target text within a 1 million-token context.
Additionally, LWM supports multimodal information processing, handling not only text but also non-textual data such as images and videos. It can watch up to an hour of video in one go and extract key information from it.
LWM’s creation marks a new stage in the development of world model technology. World models are AI models that build an understanding of the world by training on large amounts of data. LWM’s massive context window length and multimodal processing capabilities give it a more comprehensive and in-depth understanding of the world.
Researchers say LWM can be applied to a wide range of fields, including natural language processing, computer vision, and video understanding. It has the potential to drive groundbreaking advances in these areas and open up new possibilities for AI technology.
LWM is now open-source on GitHub, available for free use by researchers and developers. This will further accelerate the research and development of world model technology and inject new vitality into innovation in the AI field.
【来源】https://mp.weixin.qq.com/s/52uUGcgcoT6oGhZvi-Dl-w
Views: 1