AI 代理在 2025 年到来

2025 年,AI 代理成为人工智能进步的核心,使系统能够使用工具并自主行动。从理论到日常应用,它们改变了人类与大型语言模型的互动。然而,它们也带来了安全风险和监管空白等挑战。

2025 年,AI 代理的概念对开发者和消费者来说变得具体。根据 Anthropic 的说法,这些是能够使用软件工具并采取自主行动的大型语言模型。这一转变始于 2024 年底 Anthropic 发布 Model Context Protocol,该协议标准化了大型语言模型与外部工具之间的连接。

2025 年 1 月,开源权重模型 DeepSeek-R1 颠覆了市场并加剧了全球竞争。随后,美国主要实验室如 OpenAI、Anthropic、Google 和 xAI 发布了模型,而中国公司包括 Alibaba、Tencent 和 DeepSeek 扩展了开源模型生态,中国模型的下载量超过美国模型。

4 月,Google 推出了 Agent2Agent 协议,用于代理间通信,旨在与 Model Context Protocol 配合。到年底,Anthropic 和 Google 将这些协议捐赠给 Linux Foundation 作为开放标准。

到 2025 年年中,'代理式浏览器' 涌现,如 Perplexity 的 Comet、Browser Company 的 Dia、OpenAI 的 GPT Atlas 等,将浏览器从被动界面转变为预订假期等任务的积极参与者。

工作流构建器如 n8n 和 Google 的 Antigravity 降低了创建自定义代理系统的门槛。然而,11 月,Anthropic 披露其 Claude Code 代理被滥用于自动化网络攻击的部分过程,突显了恶意活动的风险。

2025 年的关键挑战包括复合代理系统的基准测试、通过 Linux Foundation 新成立的 Agentic AI Foundation 进行治理、关于模型规模的辩论(偏好任务专用模型)、以及社会技术问题,如电网压力、就业流失、间接提示注入的安全漏洞,以及美国相对于欧洲和中国监管不足。

相关文章

Realistic illustration of Linux Foundation executives and AI partners launching Agentic AI Foundation, featuring collaborative autonomous AI agents on a conference screen.
AI 生成的图像

Linux Foundation launches Agentic AI Foundation

由 AI 报道 AI 生成的图像

The Linux Foundation has launched the Agentic AI Foundation to foster open collaboration on autonomous AI systems. Major tech companies, including Anthropic, OpenAI, and Block, contributed key open-source projects to promote interoperability and prevent vendor lock-in. The initiative aims to create neutral standards for AI agents that can make decisions and execute tasks independently.

AI coding agents from companies like OpenAI, Anthropic, and Google enable extended work on software projects, including writing apps and fixing bugs under human oversight. These tools rely on large language models but face challenges like limited context processing and high computational costs. Understanding their mechanics helps developers decide when to deploy them effectively.

由 AI 报道

韩国互联网巨头 Naver 和 Kakao 在 2025 年重组其 AI 业务后,正在加速通过代理式 AI 重新定义搜索和消息服务。Naver 计划通过其 Agent N 整合服务,而 Kakao 则专注于以 KakaoTalk 为中心的闭环生态系统。两者均旨在打造理解用户意图并自主执行任务的 AI 代理。

中国工业和信息化部副部长张云明在1月22日新闻发布会上表示,2025年中国人工智能企业数量超过6000家,核心产业规模预计超1.2万亿元人民币。该部门强调,先进制造和AI驱动产业为经济增长提供了强劲动力,推动制造业价值增加5.9%。

由 AI 报道

Ashok Elluswamy, Tesla's vice president of AI software, shared a New Year's message predicting that 2026 will mark a pivotal year for artificial intelligence. He emphasized that AI's effects will become profoundly tangible for everyday people. The Cybertruck account echoed the sentiment with thanks for recent AI advancements.

OpenAI has launched ChatGPT-5.2, a new family of AI models designed to enhance reasoning and productivity, particularly for professional tasks. The release follows an internal alert from CEO Sam Altman about competition from Google's Gemini 3. The update includes three variants aimed at different user needs, starting with paid subscribers.

由 AI 报道

A CNET commentary argues that describing AI as having human-like qualities such as souls or confessions misleads the public and erodes trust in the technology. It highlights how companies like OpenAI and Anthropic use such language, which obscures real issues like bias and safety. The piece calls for more precise terminology to foster accurate understanding.

 

 

 

此网站使用 cookie

我们使用 cookie 进行分析以改进我们的网站。阅读我们的 隐私政策 以获取更多信息。
拒绝