文 | X研究媛 DeepSeek OCR是一个不错的OCR小模型，但过誉了。 知乎网友指出，即使横向对比OCR最优秀的几个模型，它也不是最靠前的那个。 | Model Type | Methods | Parameters | Overall | Text Idit | FormulaCDM | Table TIDS | Table TIDS-S+ | Reading Order Idit | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Pipeline Tools | Marker-1.8.2 [45] | . | 71.30 | 0.206 | 76.66 | 57.88 | 71.17 | 0.250 | | | Mineru2-pipeline [14] | . | 75,51 | 0.209 | 76.55 | 70.90 | 79.11 | 0.225 | | | PP-StructureV3 [10] | - | 86.73 | 0.073 | 85.79 | 81.68 | 89.48 | 0.073 | | Gener ...

硅谷观察/郑峻 Meta AI业务大地震！新主管上任三个月后，挥起裁员大刀，基础研究部门遭受重创，连明星大牛研究 员都不幸失业。扎克伯格这是急功近利，自毁长城；还是在精简机构，重振业务？ 前Meta AI部门员工向《硅谷观察》讲述了他开发Llama大模型的亲身经历，揭开了Meta AI部门大裁员 的背后故事。 在他看来，正是开源大模型Llama 4明显落后DeepSeek等中国竞争对手，给扎克伯格带来了强烈的危机 感，促使他下定决心、不惜血本从外部挖来AI负责人与精英人才，对Meta AI业务与战略进行彻底重 组。 Meta裁员AI部门 AI行业的竞争已经进入白热化，人才成为各家竞相争夺的最紧俏资产。谷歌和微软等诸多巨头为了集 中资源，投入AI研发竞争，在继续扩张AI研发部门的同时，甚至不惜对其他非核心部门进行大裁员。 然而，Meta一边在求贤若渴，开出天价薪酬挖角竞争对手，一边却将裁员大斧砍向自家的AI部门，开 掉了高达600名员工，甚至包括了知名的资深研究员。这让很多观察人士感到迷惑与惊讶，也让诸多AI 同行看到了招募人才的大好机会。 上周三，Meta宣布对旗下人工智能部门进行重组，裁员大约600人。这一 ...

Meta AI业务大地震！新主管上任三个月后，挥起裁员大刀，基础研究部门遭受重创，连明星大牛研究员都不幸失业。扎克伯格这是急功近利， 自毁长城；还是在精简机构，重振业务？ 前Meta AI部门员工向《硅谷观察》讲述了他开发Llama大模型的亲身经历，揭开了Meta AI部门大裁员的背后故事。 在他看来，正是开源大模型Llama 4明显落后DeepSeek等中国竞争对手，给扎克伯格带来了强烈的危机感，促使他下定决心、不惜血本从外部挖 来AI负责人与精英人才，对Meta AI业务与战略进行彻底重组。 Meta裁员AI部门 AI行业的竞争已经进入白热化，人才成为各家竞相争夺的最紧俏资产。谷歌和微软等诸多科技巨头为了集中资源，投入AI研发竞争，在继续扩张 AI研发部门的同时，甚至不惜对其他非核心部门进行大裁员。 然而，Meta一边在求贤若渴，开出天价薪酬挖角竞争对手，一边却将裁员大斧砍向自家的AI部门，开掉了高达600名员工，甚至包括了业界知名 的资深研究员。这让很多观察人士感到迷惑与惊讶，也让诸多AI同行看到了招募人才的大好机会。 上周三，Meta宣布对旗下人工智能部门进行重组，裁员大约600人。这一消息是由Met ...

Core Insights - Meta is undergoing significant restructuring in its AI department, resulting in the layoff of approximately 600 employees, including prominent researchers, as part of a strategy to enhance efficiency and competitiveness in the AI sector [3][18][21] - The layoffs are attributed to the underperformance of Meta's Llama 4 model compared to competitors like DeepSeek, prompting a sense of urgency within the company to revamp its AI strategy [10][11][21] Restructuring and Layoffs - Meta's AI department, now led by Alexandr Wang, has seen a reduction in workforce to below 3,000 employees following the layoffs [3][5] - The layoffs primarily affected three departments within the AI division, while the TBD Lab, which focuses on developing next-generation models, remains unaffected and is set to expand [5][6] - Employees affected by the layoffs were informed of their termination date, with a two-month compensation period as per California labor laws [4] Leadership Changes - Alexandr Wang was brought in to lead the AI department after Meta's significant investment in Scale AI, indicating a shift towards a more commercially driven leadership style [7][9][10] - Wang's approach emphasizes a leaner, more agile team structure, aiming to enhance decision-making and accountability within the AI division [3][10] Talent Acquisition and Competition - Despite the layoffs, Meta is aggressively recruiting top AI talent from competitors, offering substantial salaries to attract skilled professionals [6][18] - The competitive landscape for AI talent is intensifying, with other tech giants like Google and Microsoft also expanding their AI teams while cutting back in non-core areas [3][18] Impact on Research and Development - The restructuring has led to the marginalization of the FAIR research team, historically known for its foundational contributions to AI, as the focus shifts towards product-oriented development [16][17] - The integration of FAIR into the Superintelligence Lab suggests a strategic pivot towards immediate product development rather than long-term foundational research [17] Industry Reactions - The layoffs have raised eyebrows within the AI community, with many industry experts expressing concern over Meta's decision to let go of highly regarded researchers, potentially benefiting competitors [18][21] - Prominent figures from the AI field, including renowned researchers, are now seeking opportunities in other companies, highlighting the potential talent drain from Meta [21]

2025年伊始，中国楼市的下行态势愈发明显，延续着探底的脚步。根据最新统计数据，6月份全国百城二手住宅平均价格跌至每平方米13691元，环比下滑 0.75%，与去年同期相比更是下跌了7.26%。在此背景下，各地政府的"救市"行动此起彼伏，银行将房贷利率下调至3%附近，首付比例也降至15%，更有甚 者，一线城市纷纷松绑限购，如广州已明确全面取消了限购、限售及限价措施。 首先，各地房价依然存在显著的泡沫。一线城市的房价收入比高达40，而二三线城市也达到了20-25，这已远远脱离了当地居民的实际收入水平。更为严峻 的是，当前多数人的收入增长缓慢甚至出现负增长，已无力支撑高企的房价。因此，指望在未来五年内一线城市房价出现大幅上涨，并非现实之举。 其次，房屋的价值正在悄然贬值。目前一线城市中价值300万元的房产，大多属于"老破小"类型，而这类房产的抗跌性最差。以上海为例，市中心的老旧小 区，其单价已从2021年接近9万元/平方米的历史高点，回落至如今的6万多元/平方米。可以预见，一线城市的房价仍有继续下跌的空间。另一方面，在二三 线城市，300万元的房产通常意味着相当大的户型面积。这类大户型房产在未来五年内，其市值面临 ...

Market Analysis by covering: Euro US Dollar. Read 's Market Analysis on Investing.com ...

Core Insights - DeepSeek has introduced a new open-source model called DeepSeek-OCR, which utilizes a 30 billion parameter architecture to read text through images, effectively compressing text into visual tokens [1][2][19]. Group 1: Model Functionality - The model aims to replace traditional text tokens with visual tokens, achieving optical compression that allows for significant reductions in the amount of data processed [2][5]. - For instance, content that originally required 1000 tokens can now be represented with just 100 visual tokens, achieving a compression ratio of 10 times while maintaining 97% OCR accuracy [5][19]. - The model consists of two main components: DeepEncoder for image compression and DeepSeek3B-MoE for decoding the visual tokens back into text [11][12]. Group 2: Training and Data Utilization - DeepSeek trained the model on an extensive dataset of 30 million PDF documents across 100 languages, with a significant portion being in Chinese and English [12][14]. - The training also included 3 million Word documents for specialized tasks such as formula recognition and HTML table extraction, showcasing a comprehensive approach to data coverage [14][19]. Group 3: Performance and Efficiency - In tests, DeepSeek-OCR outperformed existing models like GOT-OCR2.0 and MinerU2.0, demonstrating superior performance with fewer visual tokens [16][19]. - The model's architecture allows it to operate efficiently, activating only a fraction of its parameters during processing, which enhances speed and reduces computational load [11][19]. Group 4: Philosophical Implications - The model introduces a concept of selective memory, simulating human-like forgetting by compressing older information over time, which could lead to more efficient long-term interactions [16][18]. - This approach challenges traditional notions of memory in AI, suggesting that effective information retention may not always require accumulation but rather a focus on relevance and clarity [18][22].

AI从业者正在开发应用于生活场景的AI产品，而DeepSeek带来的突破性AI模型更加提升了他们的信 心。 在人工智能(AI)这项关键性技术上，中国正处于时代前沿地位。无论是强有力的政策支持、深厚的人才 储备，还是风险资本的充足供给，都构成了我们的竞争优势。 中国在AI领域的发展也引起了投资者的广泛关注。截止到今年八月份，追踪40家香港上市AI产业链企 业的恒生人工智能主题指数上涨了34.8%，而同期反映整体市场表现的恒生指数涨幅为28.9%。 汇丰第十二届中国研讨会邀请了三位中国AI初创企业创始人一同探讨该行业的发展动态，以及他们正 在开发的AI赋能解决方案能够如何改变日常生活。 DeepSeek时刻 实际应用场景 虽然大型语言模型的发展备受瞩目，但对于企业而言，开发可实际应用的AI解决方案同样至关重要， 因为这才是实现商业化的关键。 工业机器人便是一个潜力巨大的领域。 自2021年以来，中国已连续多年成为全球最大的机器人市场， 安装量占全球总量一半以上，下一代自动化技术将很可能由AI驱动。 本代机器人 已经能够在 已知环境 中执行 预设任务 ，而 AI 的加入带来了 泛化能力——让机器人能够 在 陌生环境 ...

来源：21世纪商业评论 本文字数：1777｜预计3分钟读完 各显神通，胜负未分。 作者丨子川 AI圈和投资圈，最近都被一场炒股大赛刷屏！ 一家名叫Nof1的初创公司，组织了一场AI炒股大赛——Alpha Arena。 而收益率最低的居然是用户量最多的GPT-5，目前账户还剩5414美金，亏损高达45.81%，属实是亏麻了！ 纵观整个收益曲线，GPT-5只有在比赛开始的时候小有盈利，其他时间都在亏，特别是19号以后，收益曲线一降再降，根本就止不住。 收益率仅次于DeepSeek的是Grok 4，目前收益是11.7%，此前最高的收益率是38.15%。 和GPT-5相反的是，除了在第一天亏了点外，其余时间都在盈利，曾经有段时间收益率一度接近DeepSeek。 这并非是模拟盘，而是给每个AI工具1万美元，让它们自己炒股，并对收益进行24小时监控。 此次比赛从美东时间10月18号开始，为期2周，在11月3号结束，目前是第四天。 1 顶级阵型 参赛的AI选手阵容堪称豪华，汇集了国内外最顶级的AI模型。 海外组： OpenAI的GPT-5、谷歌的Gemini 2.5 Pro、马斯克的Grok 4，以及Anthropic的C ...

谁是最赚钱的大模型？给六个顶级AI各发1万美刀的启动资金，让它们直接在真实市场里操作，看谁活 得更久，赚得更多？ ...