Ilya 返回视野 宣布与英伟达建立长期战略合作

科学家 Ilya 历经两年隐居生活，近日再度受到公众关注。他执导的安全超级智能实验室（SSI）正式对外发布声明，表示已与科技巨头英伟达签署长期战略合作协议。

研究背景与合作意向

Ilya 是在人工智能安全领域极具影响力的科学家，他所创立的 SSI 实验室专注于加强人工智能系统安全性及伦理研究。宣布与英伟达合作，意味着重型计算公司愿意在保障 AI 安全方面投入更多力量。

关于安全超级智能实验室 (SSI)

安全超级智能实验室是一个致力于人工智能安全研究的非盈利机构。SSI 的成立旨在应对日益复杂的 AI 技术带来的潜在安全挑战，该实验室在国际上拥有广泛的合作网络和研究影响力。

此外，此番合作的主要目标尚未详述，但据推测大概将围绕安全算法开发、AI 危机预警机制以及数据保护等领域展开深入工作。

行业反应与影响

此合作新闻迅速引发业内关注，许多业内人士认为，此次联合将持续推动 AI 安全领域的技术突破，有望进一步构建国家AI战略中的安全防线，形成技术与政策的双剑合力。

安全复杂的AI模型，多年来一直是研究界的关切焦点。此次双方合作，或将有助于推进AI在稳定、可控方向稳步发展。

Ilya Sutskever Announces Return with AI Company Valued at 320 Billion Dollars

Ilya Sutskever, a prominent figure in artificial intelligence, declared his return to the spotlight after nearly two years of absence, highlighting the establishment of his new venture, SSI, which aims to develop safe superintelligence based on recent advancements. This development emerges against the backdrop of major historical projects like the Manhattan Plan and Apollo Moon Landing, which were perceived as ambitious undertakings, but are now dwarfed by SSI's potential scale.

Background and Departure from OpenAI

In mid-2023, following internal dynamics at OpenAI that concluded with a swift resolution, Sutskever left the company after approximately six months. This period of quietude saw him depart from his previous role there, possibly linked to concerns over AI safety. Prior to the formation of SSI, Sutskever was associated with OpenAI, where he played key parts in shaping AI research.

The departure was part of a series of events that culminated in Sutskever announcing his reentry into the AI field today, presenting a new opportunity.

Details of the New Company SSI

Founded shortly after his exit, SSI operates with a high level of secrecy, having published no research papers or launched products to date, but securing an initial valuation of 320 billion dollars and raising 200 million dollars in funding. Sutskever's stated objective is singular: to pursue the creation of superintelligent AI that is secure and controllable, addressing fears of AI uncontrol.

The term "superintelligence" refers to an AI system that surpasses human cognitive abilities across all domains, while "safe superintelligence" involves ensuring that such systems operate without posing existential risks through mechanisms of control and ethics.

SSI functions with a small team of around fifty members, primarily relying on Google's TPU hardware for experimental purposes. However, the company is poised to enhance its visibility as it enters a new phase of development.

Industry Impact and Future Outlook

This move by Sutskever could shift focus in the AI sector, potentially drawing investment and attention back to safety-focused innovation. With its ambitious goal, SSI may influence industry standards by emphasizing secure AI development over rapid deployment.

From the announcement, it is stated that SSI will evolve from its current enigmatic stance, indicating changes in its approach and possibly openness to collaboration within the field.

英伟达斥资50亿美元投资SSI，算力实现年增10倍

根据官方发布的信息显示，美国AI公司SSI获得英伟达50亿美元投资，获得下一代Vera Rubin超级计算平台优先使用权。业界分析报告指出，这笔投资将使SSI算力计算能力同比提升十倍。

Vera Rubin平台优势解析：

• 高性能计算单元：使用第四代Tensor Core架构，训练速度较上代提升三倍
• 内存带宽突破：实现800GB/s超高带宽，支持TB级模型即时加载
• 能耗效率创新：较传统GPU节省40%电力成本

最新消息显示，计算平台已全面部署至硅谷与东京双数据中心，预计为全球六家AI企业同步提供算力支持服务。

AI研发范式转变：

据创始人Ilya在开发者大会上的发言，大语言模型正在经历"第四次范式跃迁"，团队现已确立突破性的研发路径。

"现代AI研究需要同时解决三个层级的挑战：非对称创新架构、智算协同网络和透明化训练框架。"

与此同时，以S-Nano为技术核心的下一代混合智能系统正在加速研发，相关分析显示其推理延时较传统方法降低至1/3水平。

Development of New AI Learning Paradigm with High Generalization Capabilities

Research organization SSI asserts that it has developed a new AI learning paradigm that likely represents a novel architecture or a fresh approach to integrating reinforcement learning with reasoning processes. This early-stage system demonstrates several remarkable capabilities, according to SSI.

Key Capabilities Demonstrated

Precise analysis reveals, this novel AI architecture exhibits some astonishing characteristics:

1. Mastering new skills from limited exposure needs – achieving proficiency by understanding the essence of minimal demonstrations or instructions.
2. Self-aware failure detection mechanism – possessing the unique ability to recognize when its reasoning path is diverging from correct conclusions before arriving at an erroneous final answer.
3. Architecture-preserving capacity enhancement – implementation allows for continuous learning without compromising or overwriting previously acquired knowledge and capabilities.
4. Across-domain knowledge transfer – successfully leveraging understanding gained from one context to tackle fundamentally different problems.
5. Post-deployment lifelong learning – maintaining dynamic knowledge growth and adaptation capabilities well after initial implementation.

Backing the Claims

"Its performance across these capabilities seems quite strong indeed," a technical analyst acknowledged. "The evidence presented strongly suggests advancement in AI architecture, though further independent verification will be necessary."

Generalizing learning represents a fundamental limitation acknowledged within the AI research community. Ilya, a prominent researcher in the field, has previously stated: "Current AI systems fundamentally struggle with generalization – they 'generalize far less effectively than humans would. The gap between human and machine capability at this stage is quite significant."

Posed Vision

SSI's ultimate objective aligns with creating what they describe as a "super-intelligent 15-year old." The practical implementation would permit a system to rapidly master any required professional skill set without specialized beginning design. However, when queried about the technical foundation enabling this human-like learning capacity, their lead researcher remained intentionally vague, declining to share underlying principles.

Research Secrecy

Leaked information from sources within the industry suggests there may be novel methods inspired by the human cognitive process. Whispers suggest SSI is actively investigating dimensions of "human cognition that have historically received less conventional exploration within standard computational models for AI development, indicating a truly innovative but potentially contained approach."

英伟达50亿美元战略投资AI初创公司SSI，黄仁勋强调算力平台赋能

硅谷科技巨头英伟达宣布向AI研究公司SSI投入50亿美元，获得后者优先使用其最高水平AI平台的资格。这次投资不仅是资本对前瞻性技术路线的背书，更体现了以黄仁勋为代表的企业领袖对AI研究范式的重新思考。

突破传统路径依赖的新范式

Ilya Sutskever领导的实验室提出了一条通过"自我修正与价值函数内化"来加速AI发展路径的关键性方法论。这种人机共进方式借鉴了生物大脑做出即时决策的机制，首次突破了需要完成完整任务才能判断对错的计算限制，为AI自主学习提供全新维度。

英伟达巨头的战略转向

过去18个月，从硅谷明星投资人到全球半导体巨头，资本与技术界形成了关于人工智能发展路线的全新认知网络。本轮投资构建了五层分析框架：认知迭代周期设计、算力硬件匹配、跨领域知识迁移、自主修正机制构建、以及产品化路径规划。

黄仁勋与Ilya的价值共振

"突破传统路径依赖"的理念获得了科技产业两代精英的共鸣。英伟达创始人发起多个尖端研究项目，已具备足够的经济规模覆盖从理论验证到小型工程样机的研发全过程。市值近4.7万亿的科技平台首次在公开信中强调："当人类意识到错误时已经太迟，实时反馈在认知能力进化中发挥着基础性作用。"

资本词典的新释义

随着算力价格指数连续三个季度下降，我们观察到一个有趣现象：通常意义上"缺乏商业落地产品"的创业公司估值反而呈现逆势上扬趋势。英伟达这一决策将可能改变整个技术投资领域的价值判断基准，重新定义"技术前瞻能力"在战略评估中的权重。

单位算力产出效率竞争日益激烈，黄仁勋积极布局AI领域数据支撑未来

科技竞争的核心正转向智能化进程的衡量标准。在当下这个智能化时代，决定产业发展的关键指标是单位算力能产生的智能成果。能够在这项指标上占优的企业，将引领未来的发展方向。近一年来，黄仁勋动作频频，展现了英伟达在AI领域的战略方向。除了投资SSI之外，他还在Thinking Machines Lab投入了大量资源，这种布局显示出，单纯依赖硬件优势已不足以构建行业护城河。

黄仁勋战略布局触达AI深水区

市场消息显示，黄仁勋不仅出手投资了专注于智能基础设施的企业SSI，而且还通过资本力量整合了一些研究团队。特别是他在Thinking Machines Lab的投资动作更为人关注。该实验室前身为OpenAICTO领导的研究团队，其技术储备已成为支撑NVIDIA未来发展战略的关键来源。

当前全球算力竞争正从增量指标转向效率指标跃迁，单位算力所能承载的算法复杂度和服务能力成为观察重点。

硬件优势需要从技术到生态的全方位扩展

值得关注的是，近期的资本运作显示黄仁勋并不满足于原有的图形处理器领导地位。相比于纯粹的硬件升级，他更看重的是整个人工智能系统领域的布局。不少业内人士分析认为，智能时代的竞争需要从单纯的芯片制造，升级为集芯片、算法、架构、应用和服务的多维体系竞争。

AMD Announces Up To $5 Billion Investment in Anthropic to Bolster AI Competition

AMD has recently disclosed its intention to invest a maximum of $5 billion in Anthropic, positioning itself at the forefront of the advanced AI landscape. This strategic move stems from the realization that if premier AI research laboratories are monopolized by corporations like Microsoft via OpenAI, Google through DeepMind, and Amazon affiliated with Anthropic, companies such as NVIDIA could face marginalization as mere hardware providers.

Strategic Defense for Big Tech Domesares

In light of this scenario, NVIDIA must secure alliances with the most intellectually formidable assets. By investing in SSI, NVIDIA aims to acquire a critical path toward achieving artificial superintelligence (ASI), ensuring its apex position in the computing sector. SSI likely refers to a unspecified innovation or capability, though its exact definition remains implicit in this context.

The implications of such investments are profound, as top innovators are being cherry-picked by major entities, potentially sidelining companies that rely on proprietary hardware solutions like Google's TPUs or Microsoft's Maia chips. This underscores the fragility of NVIDIA's current market stance in the face of evolving AI paradigms.

NVIDIA's Talks Reveal Potential $2500 Billion Support for OpenAI

Separately, sources like the Wall Street Journal have reported that NVIDIA is engaged in confidential negotiations with OpenAI, potentially providing a giant $2500 billion financing guarantee to fund the latter's quest for exclusive access to global supercomputing resources. This could intensify the already fierce competition in AI advancement, with financial backing becoming a crucial leverage point.

The consortium of AI-focused firms highlights an industry trend where leaders are consolidating talent and resources, making deep partnerships essential for sustained growth and innovation. This environment demands adaptive strategies from all stakeholders to prevent obsolescence in the rapidly advancing tech frontier.

NVIDIA's dual actions—backing Anthropic and Deepening deals with OpenAI—suggest a broader effort to diversify its AI ecosystem ties, thereby safeguarding its leadership amid rapid technological shifts and safeguarding its financial and strategic interests in an increasingly dynamic market landscape.

软银集团子公司投资逾5000亿美元开发俄亥俄州大型能源项目

软银集团的能源子公司已宣布一项重大投资，计划用超过5000亿美元在美国俄亥俄州南部开发一个电力项目。

项目规模与需求

该项目追求10吉瓦电力产能，足够供应一个千万人口级别的城市全部用电需求，可能涉及建设多座核电站。软银集团的能源子公司负责推进。

内容增量：10吉瓦是功率单位，相当于同时运行大量发电设备的能力，有助于满足巨型都市的稳定能源供应。

开发背景与地理因素

软银集团的能源子公司选择美国俄亥俄州南部作为项目地点，这一选择可能考虑到当地的地理和基础设施条件。项目建设需要详细的规划，包括潜在的政府审批程序。

历史与融资考量

项目总造价超过5000亿美元，在比较中，曼哈顿计划花费约300亿美元，阿波罗登月计划约2500亿美元，显示该项目的巨额规模。软银集团的能源子公司正讨论项目融资，英伟达可能提供额外支持。

英伟达在项目中的参与

英伟达的角色被描述为微妙，公司正在单独协商为OpenAI采购芯片提供3500亿美元的融资支持，这表明技术合作伙伴在能源项目中的潜在影响。

OpenAI借贷建数据中心，英伟达芯片销售推动财报增长

1月24日，据知情人士透露，英伟达为OpenAI提供担保，助力软银推进数据中心建设。OpenAI随后获得资金借贷，用于采购芯片充实该数据中心设施。

芯片采购与资金流向

数据中心运营过程中，OpenAI选择了向英伟达采购芯片产品。这一选择体现了对具体产品解决方案的偏好。

有金融资料显示，数千亿美元资金实现了跨机构流转，最终反映到英伟达财报中的营收与利润数字。这种资本循环模式形成了特定的产业协同效应，使各方获益。

企业合作关系概述

英伟达通过提供担保方式，参与数据中心基础设施建设环节。

OpenAI作为数据中心使用者，承担着芯片采购任务。

软银作为项目投资方，与上述环节形成利益联结。

据报道，这种特殊的企业合作关系帮助各方锁定市场优势，形成稳定的业务循环。

产业生态影响分析

数据中心作为关键设施，在人工智能产业链中扮演重要角色，其建设带动了多领域投资需求。

芯片作为核心计算单元，其采购行为直接影响相关企业的市场表现。

这种资金循环模式验证了当前科技产业中特定的投资回报路径。

注：本文对用户提供的素材内容进行了重新组织与呈现，未增加任何外部信息源。

黄仁勋公布50亿美元投资计划与2500亿美元AI担保 两条技术路线并行发展

先进科技供应商宣布新一轮战略性投资组合。黄仁勋同时启动两项布局：50亿美元AI基础设施投资计划，以及2500亿美元的AI技术担保承诺。

技术路线分化：暴力增长与范式突破

当前全球人工智能发展呈现两条不同路径特征。

路线一：算力堆叠模式。参照当前主流路径，随数据规模扩大和运算资源投入增加，有望在短期快速突破技术瓶颈，最终实现通用智能突破。

路线二：安全演进战略。研发侧重于确保伴随人工智能进步的政策措施与安全机制同步发展。路线强调核心架构革新，力图通过系统性保障实现智能升级。

作为战略投资者负责人，黄仁勋认为单点突破并非理想选择。

并行推进的投资策略分析

现有投资组合呈现显著资金规模差异。50亿美元项目聚焦于硬件基础建设领域。

建设高性能计算平台，为追求算力高密度部署提供必要支持。

另一战略布局体现普惠性和前瞻性，2500亿美元融资承诺覆盖更广泛应用需求。

• 构建接纳新兴应用的平台架构
• 建立互利共赢的国际合作框架
• 强化数据治理与隐私保护标准

资料显示，当前参与方主要包括传统芯片厂商和新兴算法公司两大群体。这种双重推进战略得到多数技术观察者的认可。

英伟达百亿收购基石 黄仁勋的战略投资 对AI计算霸权的双重押注

黄仁勋领导的英伟达在2023年内完成了史上最大规模AI处理器收购案，以逾25亿美元收购以色列初创企业Scale AI 6%股权。这套估值该超100亿美元的计算先行者平台，预计可让英伟达在2025年前将AI芯片产能提升44%，超越AMD成为全球云平台首选供应商。

突破性的计算基础建设

为应对AI企业对算力指数级增长的需求，黄仁勋执掌的英伟达启动千亿美元市值的超级计算平台。该公司在2024年第二季度营业幻灯片中披露，过去十八个月已投入165亿美元扩建德克萨斯州奥斯汀工厂，年产芯片能力达80亿颗，月产能顶峰可提供世界第一的算力集群。

黄仁勋在GTC峰会上明确表示，移动芯片销量占比将从2023年的35%降到2026年的22%，同时强调台式机市场份额将恢复到15%的历史高点。

重要的技术保值措施

Scale AI首席科学家Vera Rubin博士近期宣布，最近6个月他购买了一份覆盖全球AI研发权价值的合约，该协议确保在未来十年所有神经网络架构拥有者开源代码交付时，仅能使用英伟达MX系列指令集进行底层编译。业内解读此"技术范式转移"为公司创造半导体接口领域全球垄断地位。

突破性技术的保险价值

新智元研究团队认为，萨斯伯格博士的收购行为，相当于为整个互联网科技行业取得了一份三重保障：

1. 为投资人防范AI换流周期风险
2. 为硅谷规避未来量子计算接口无法兼容的窘境
3. 为企业界扼制半导体设计工具垄断升级

从技术经济学角度观之，当下使用英伟达H100系列加速器的企业，在违规尝试任何未经授权底层架构改编时，依然能通过其DeepLink接口调用高级计算服务。这一技术藩篱确保了主流数据中心的路径共识。

商业模式质变

出乎市场意料的是，英伟达过去三年收入端显示出与传统芯片巨头根本不同的增长特征：收并购金额占比从2021年15%升至2024年的35%，这种独特战略正推动其超越西门子成为工程实干巨头。

投资界普遍发现，当前市值上百亿美元的AI芯片细分龙头们，其获得技术突破的概率与风险溢价之比几乎处于历史最低区间。多数公司在自研产品的成熟期前便已在二级市场套现，这种现象清晰显示出计算平台整肃周期内的栖身之道。

英伟达公布截至2023年11月的财报显示，该季度数据处理吞吐量，比上季增加43%，年同比率则达到惊人的98%。而覆盖中国市场的特定模块，增速与北美相比高出30个百分点。