贝特森只是一群重要AI研究人员和高管中的一员，他们正面临无休止的高强度工作，为了追求超越人 类智慧的AI系统，不断与看似永无止境的技术变革赛跑。 100小时 凤凰网科技讯 北京时间10月23日，据《华尔街日报》报道，AI行业的突飞猛进使得硅谷巨头们不敢有 丝毫放松。虽然AI研究员已是各大公司疯抢的人才，但他们面临高强度工作，即便薪酬不菲，也没时 间花。 AI研究员乔希·贝特森(Josh Batson)如今已无暇浏览社交媒体。现在，唯一能让他感到愉悦的就是 Anthropic公司的Slack工作频道。在那里，他沉浸于同事们关于大语言模型与架构的理论探讨和实验交 流。 在硅谷最顶级的AI实验室内部，顶尖研究员和高管们每周通常要工作80小时至100小时。多位顶尖研究 人员将这种工作环境比作战场。 "我们实际上是在尝试用两年时间快速完成本该二十年实现的科研进展，"Anthropic研究科学家贝特森表 示。他指出，AI系统"每隔几个月"就能实现突破性进展，"它是当今世界上最有趣的科学问题"。 微软、Anthropic、谷歌、Meta、苹果和OpenAI的高管与研究人员都表示，他们正身处历史分水岭，既 要与对手展开角逐， ...

Core Insights - The quantum technology sector experienced a significant surge, with companies like Geer Software hitting the daily limit up, and others such as Jida Zhengyuan, Hexin Instruments, Keda Guochuang, and Guodun Quantum also seeing gains [1] Group 1 - Google's announcement on October 22 regarding its "Willow" quantum chip achieving the first verifiable quantum advantage algorithm on hardware has driven interest in the quantum technology sector [1] - The performance of the "Willow" quantum chip surpasses the fastest classical supercomputers by a factor of 13,000 times, highlighting its potential impact on computational capabilities [1]

Market Overview - Despite high tariffs and a 22% correction in the S&P 500 earlier this year, equities are near all-time highs entering Q4, supported by monetary policy easing [2] - The Federal Reserve cut short-term interest rates to 4.00%–4.25% on September 17, indicating progress on inflation and softer labor conditions [2] - Markets anticipate two more 25 basis point cuts by year-end, contingent on cooling core service and wage inflation [2] Inflation and Consumer Impact - The headline consumer price index (CPI) is up 2.9% year-over-year, with low-income consumers feeling strain while high-income consumers remain resilient [5] - Goods deflation and cheaper traded inputs have mitigated the impact of tariffs on everyday prices, with import prices remaining flat to down through mid-2025 [4] - Core PCE inflation is in the high-2s, with stickiness in services rather than tariff-exposed goods [4] Corporate Activity and M&A Trends - Corporate boardrooms are increasingly engaging in mergers and acquisitions, driven by easing funding costs and a pursuit of scale [6] - Valuations have re-accelerated despite mixed deal volumes, with expectations for continued M&A activity in AI-adjacent tech, infrastructure, and select industrials [6] Earnings and Valuation Insights - The S&P 500 is near all-time highs with a forward 12-month price-to-earnings ratio of 22–22.5x, above historical averages, limiting expansion of stock valuation multiples [7] - Continued profit growth and free cash flow durability are essential for the next leg up in the market [7] AI Infrastructure and Investment Dynamics - Corporate investment in AI is driving market dynamics, with capital expenditure extending beyond GPUs to the entire infrastructure stack [11] - OpenAI is central to this investment shift, leveraging its user base to influence the AI value chain [12] - OpenAI's partnerships and contracts, including a reported ~$300 billion deal with Oracle, indicate a shift towards debt-fueled funding in the AI sector [16] Company-Specific Insights: Ferguson Plc - Ferguson is the largest specialty distributor for North American plumbing, with a revenue split of ~51% residential and 49% non-residential [22] - Despite a 16% drop in shares post-earnings due to fears of commodity deflation, revenue held steady, indicating resilience in pricing power [23] - The company is expected to continue compounding growth through organic means and accretive M&A, benefiting from structural advantages in sourcing and efficiency [25] Company-Specific Insights: Broadcom - Broadcom has been a strong performer in the semiconductor sector, positioned as a key player in the AI market alongside Nvidia [27] - The company excels in custom AI chips and networking solutions, with significant revenue growth expected in its AI segment [29] - Broadcom's strategic M&A and strong balance sheet position it well for future growth, particularly in AI and networking [33]

今天我们继续聚焦量子计算。 上篇我们详细探讨了量子计算的技术原理、商业化瓶颈、以及量子纠错领域的前沿进展。 下篇主要聚焦这么几个问题： 量子计算的理解门槛较高，内容相对硬核，建议耐心阅读。 （1）经典计算的瓶颈与量子计算的优势 随着制程逼近物理极限，经典计算的天花板已经近在眼前，量子计算呼之欲出。 首先是计算瓶颈。经典计算的算力是线性增长的，而量子计算由于"量子叠加"原理，可以实现算力的指数级增长。 当遇到如药物分子模拟这类复杂度较高的问题时，即使用当下性能最强大的超算，也需要花费数亿年的时间才能验算完成，量子计算则可以快速解决问 题。 经典计算的第二个问题是量子隧穿现象。 当电子元件缩小到纳米尺度，绝缘层薄得只剩几个原子，此时电子会出现"隧穿现象"，穿墙而过导致漏电，使得晶体管失灵，摩尔定律撞上了硬邦邦的量 子墙。 在超导量子计算中，结成库珀对的电子采取量子隧穿的方式通过约瑟夫森结，因此不存在这方面的瓶颈。 | 时间 | 战略/规划法案 | 国家/地区 | 投资规模(美元) | | --- | --- | --- | --- | | 2014 | 国家量子技术计划 | 英国 | 10年投资约12.15亿美元 ...

Group 1: Core Value of Quantum Computing - The value of quantum computing lies not in replacing existing computers but in its ability to solve problems that classical computers cannot, thereby creating new markets [1] - Investment opportunities in this field should prioritize technological strength, particularly the path to "fault-tolerant" computing, moving beyond merely increasing the number of physical qubits [1] Group 2: Google's Willow Chip - Google's Willow chip, set to be released in December 2024, demonstrates the scalability of quantum error correction, addressing a significant challenge in the field for nearly 30 years [2] - The achievement shows that as the scale of encoding increases, the logical error rate decreases exponentially, providing a clear experimental path for building large-scale, reliable fault-tolerant quantum computers [2] - Google showcased "quantum supremacy" by completing a computation in under 5 minutes that would take classical computers 1,025 years [2] Group 3: Applications of Quantum Computing - Future quantum computers will work alongside classical computers to form new supercomputing architectures, focusing on four core areas: 1) Quantum simulation for drug discovery and materials science, enabling unprecedented precision in simulating molecular behavior 2) Combinatorial optimization for finance and logistics to find optimal solutions among vast possibilities 3) Empowering artificial intelligence by processing complex models and high-dimensional data, potentially leading to exponential acceleration in machine learning 4) Algorithm-defined advantages in specific fields like cryptography using algorithms such as Shor's [3] Group 4: Technical Routes and Key Companies - The hardware solutions in quantum computing have not yet converged, with major players like Google and IBM advancing the superconducting route, leveraging breakthroughs in quantum error correction [4] - Companies like Rigetti and domestic firm Benyuan Quantum are agile challengers in the same field due to their unique chip manufacturing capabilities [4] - The "quality over quantity" philosophy has led to the emergence of high-potential paths, such as IonQ's ion trap technology, which boasts near-perfect qubit fidelity and full connectivity [4] - Other companies like Infleqtion (neutral quantum bits) and D-Wave (quantum annealing) are building unique technological barriers in their respective niches [4]

谷歌公司Quantum周三在《自然》杂志披露与Willow芯片相关的量子计算突破性进展，该公司Willow量子计算芯片上运行了一种算法，这种算法可以在类似的 平台上重复运行，并且性能超越传统超级计算机。 谷歌表示，这一突破为量子技术在未来五年内实现实际应用铺平了道路。 谷歌表示，这项被称为"量子回声（Quantum Echoes）"的算法具有"可验证性"，意味着可以在另一台量子计算机上重复运行。谷歌强调，该算法的运行速度 比全球最强的超级计算机快1.3万倍。 谷歌称，这些成果结合在一起，意味着量子计算在医学和材料科学等领域将有广泛的潜在用途。 谷歌量子AI部门的研究科学家汤姆·奥布莱恩（Tom O'Brien）表示： "可验证性这一点非常关键，这是我们迈向现实世界应用的重要一步。实现这一成果意味着我们正在真正推动量子计算走向主流。" 媒体称，这项突破使谷歌更接近真正利用量子计算所承诺的强大运算能力。目前，微软（Microsoft）、IBM以及众多初创企业也正在该领域积极追赶。 就在去年12月，谷歌曾宣布"Willow"芯片仅用五分钟就解决了一个问题，而同样的问题若由超级计算机处理，需要10秭（10 septi ...

国际最新研究："逆转时间"或能探测量子计算机特性并提升其性能 中新网北京10月23日电 (记者 孙自法)国际知名学术期刊《自然》最新发表一篇量子物理论文称，研究 人员开展一项"逆转时间"探测量子动态的研究提出，通过逆转信息置乱的方式操控量子回路，或能探测 量子计算机的特性并提升其性能。 这项研究由谷歌量子AI与合作者团队完成，他们在论文中报告了在超导量子处理器中测量非时序关联 子(OTOC)的量，OTOC可作为理解量子计算机的工具，也即一种研究量子信息如何在多粒子量子系统 中传播的工具，用于构建超越经典计算机性能的可验证演示。 来源：中国新闻网 编辑：付健青 广告等商务合作，请点击这里 本项研究成果相关示意图(图片来自论文)。施普林格·自然 供图 论文作者团队介绍，量子计算长期以来的目标，是打造性能足以实现量子优势的量子计算机，在特定和 理想的实用任务中超越经典计算机。要实现这一目标，需要通过降低噪声和缺陷克服来解决一系列的难 题；其中一个问题是探测系统中众多组件的量子动力学，以区分真实量子效应和经典噪声。 他们指出，这些系统可能难以研究，因为相互作用要素的行为不可预测且难以经时追踪，尤其是仅于特 定时刻测量 ...

| 机器之心报道 | | --- | | 机器之心编辑部 | 刚刚，谷歌宣布了一项具有历史意义的研究成果。 他们全新的量子回声（Quantum Echoes ） 算法在 Willow 芯片上运行，解决原子相互作用问题的速度比最好的传统超级计算机快 13000 倍，在数小时内完成了需要 Frontier 超级计算机大约 3.2 年才能完成的计算。 更令人称奇的是，其结果是可验证的，可以说这是量子计算机首次能够在真实硬件上成功运行可验证的算法。相关研究登上 Nature 封面。 早在 2019 年，谷歌就曾展示过量子计算机能够解决一个经典超级计算机需要数千年才能完成的问题。而在去年年底（2024 年），谷歌推出的新一代 Willow 量子 芯片展示了如何显著抑制误差，解决了科学家们近 30 年来一直面临的核心难题。如今的这一突破，让谷歌量子计算机朝实用性发展，又迈进了一大步。 这项工作涉及谷歌量子 AI 团队的许多成员，以及谷歌 DeepMind 和加州大学伯克利分校、达特茅斯学院等研究者。值得一提的是，新晋诺奖得主、现任谷歌量子 AI 实验室硬件首席科学家 Michel Devoret 也参与其中。 量子回声核 ...

Core Insights - The favorable antitrust ruling for Alphabet's Google has allowed the company to maintain its core search business, positively impacting its stock performance and contributing 3.1% to the fund this quarter [3] - UnitedHealth's stock rebounded after a significant decline earlier in the year, adding 0.9% to the fund, driven by optimistic comments from the new CEO and investment interest from Berkshire Hathaway [4] - Progressive was the largest detractor in the quarter, reducing performance by 0.5% due to concerns over lower interest rates affecting investment income [5] Performance Summary - The Bretton Fund achieved a return of 8.21% for the third quarter, outperforming the S&P 500 Index, which returned 8.12% [7] - Over the past year, the fund's return was 8.92%, while the S&P 500 Index returned 17.60% [7] - The fund's inception date was September 30, 2010, and it has delivered a 12.89% return since inception [7] Portfolio Composition - As of September 30, 2025, Alphabet Inc. constituted 11.51% of the fund's net assets, making it the largest holding [11] - Other significant holdings included AutoZone Inc. at 7.11% and The Progressive Corporation at 6.84% [11] - The fund also held positions in major companies like American Express, JPMorgan Chase, and UnitedHealth, each contributing to the overall portfolio diversity [11] Investment Actions - The fund sold its position in Union Pacific Corp after nearly 15 years, achieving a 13% annualized internal rate of return [14] - The decision to sell was influenced by concerns over a pending acquisition that could dilute shareholder value and distract management from core operations [15]

2025.10. 23 图片来源：谷歌博客 量子计算自上世纪80年代被提出以来，一直是科技圈关注的焦点，过去几年来，包括谷歌、微软， IBM等大厂都在研发量子计算机。早在2019年，谷歌曾在《Nature》发布了一项量子计算突破，在世 界第一超级计算机Summit需要计算1万年的实验中，谷歌研究的量子计算机只用了短短3分20秒。当 时，谷歌CEO桑达尔·皮查伊（Sundar Pichai）将此次突破比作"莱特兄弟的首飞"。 本文字数：1633，阅读时长大约3分钟 作者 | 第一财经 刘佳 谷歌当地时间周三在《自然》杂志披露与Willow芯片相关的量子计算突破性研究成果。该公司称这是 历史上首次证明量子计算机可在硬件上成功运行一项可验证算法，其性能甚至超越了最快的经典超级计 算机（速度快 13000 倍）。 Willow是谷歌于去年12月宣布推出的量子芯片。当时，Willow量子芯片在5分钟内完成了一项传统超级 计算机需要"10的25次方"年的时间才能完成的标准基准计算任务。而此次谷歌披露量子可验证性意味 着，该结果可在谷歌的量子计算机（或其他同等水平的量子计算机）上重复得出，从而确认结果的准确 性。这种可重复 ...