In January 2025, DeepSeek released its latest open-source AI model, DeepSeek-R1, which challenges the long-standing myth in AI development that "bigger is always better." This model achieves performance close to that of OpenAI's GPT-01 at a fraction of the cost, sparking extensive discussions in the industry. Experts have noted that DeepSeek's success could potentially ignite a wave of innovation and drive global progress in AI technology.

Sam Altman, CEO of OpenAI, acknowledged that DeepSeek-R1 is an impressive model, especially considering its cost-effectiveness. Peter Wennink, CEO of Dutch semiconductor equipment manufacturer ASML, highlighted that the launch of an efficient AI model by DeepSeek is good news for the chip market, as it helps reduce the cost of AI applications.

Morgan Stanley, an international investment bank, pointed out that DeepSeek has demonstrated an efficient training approach different from traditional "large models" by improving model architecture and data quality. This marks a shift from the notion that "bigger always means smarter." The "Financial Times" also commented that the open-source model DeepSeek-R1 has attracted significant global interest, which is conducive to the development and application of AI technology.

On January 20th, 2025, US President Donald Trump, along with OpenAI's CEO Sam Altman and SoftBank's chairman Masayoshi Son, announced the launch of the "Stargate" project, which plans to invest 100 billion dollars to build data centers in the US to support AI development. Although Trump called it the "largest AI investment project in history," the funding sources for the project have been questioned. Additionally, the US magazine "Information" revealed that OpenAI and Microsoft had already planned a similar data center project in Texas.

As AI technology continues to mature, countries around the world are paying close attention to its potential and risks. UN Secretary-General Guterres emphasized at a conference that the unregulated expansion of AI could pose global risks and called for strengthened international cooperation to ensure that AI development benefits humanity as a whole.

Source: Lingnan on the Cloud

DeepSeek捅穿了“星际之门”?

2025年1月，深度求索发布了最新开源AI模型DeepSeek-R1，打破了人工智能开发中的“越大越好”迷思。这一模型以低成本接近OpenAI的GPT-01性能，引发业界热议。多位专家表示，DeepSeek的成功或将激发创新浪潮，推动全球AI技术进步。

OpenAI首席执行官萨姆·奥尔特曼称，DeepSeek-R1是一款令人印象深刻的模型，尤其考虑到其在同价位的表现。荷兰半导体设备制造商阿斯麦总裁兼首席执行官富凯表示，对芯片市场来说，深度求索推出高效AI模型是个好消息，有助于降低AI应用成本。

国际投行摩根士丹利也指出，DeepSeek通过改进模型架构和数据质量，展现了一种与传统“大模型”不同的高效训练方式，标志着“更大不等于更聪明”。此外，英国《金融时报》评论文章说，开源模型DeepSeek－R1对全球用户产生极大吸引力，有利于推动人工智能技术的开发和应用。

1月20日，美国总统特朗普与OpenAI首席执行官奥尔特曼、软银董事长孙正义等宣布启动“星际之门”项目，计划投资1000亿美元在美国建设数据中心，支持AI发展。尽管特朗普称这是“史上最大”的AI投资项目，但项目资金来源引发了争议，而美国《信息》杂志则披露，OpenAI与微软早已有类似计划在得州建设数据中心。

随着AI技术不断成熟，全球各国纷纷关注其潜力与风险。联合国秘书长古特雷斯在年会上强调，AI的无序扩张可能带来全球风险，呼吁加强国际合作，确保人工智能的发展惠及全人类。

文｜来源于新华社
译｜赵凡
英文审校｜林佳岱