1,648 Matching Annotations
  1. May 2026
    1. Here, he’s been consistent; in March 2024 Karp told a CNBC reporter that ‘if you have a position that does not cost you ever to lose an employee, it’s not a position’

      This statement by Alex Karp suggests a focus on employee turnover as a measure of company health, which may require further analysis of his management style.

    2. Karp gave an interview to CNBC claiming that AI could undermine the power of ‘humanities-trained—largely Democratic—voters’ and increase the power of working-class male voters

      This statement by Alex Karp is a non-consensus view on the impact of AI, which may require further analysis of its implications and potential biases.

    3. At one point during the call, one of the employees tried to level with the group, explaining that Palantir’s work with ICE was a priority for Karp and something that likely wouldn’t change any time soon.

      This statement indicates a high priority given to Palantir's work with ICE by the CEO, which may be a point of contention among employees.

    4. Last fall, Palantir seemed to become the technological backbone of Trump’s immigration enforcement machinery, providing software identifying, tracking, and helping deport immigrants on behalf of the Department of Homeland Security

      This statement suggests a significant role of Palantir in Trump's immigration enforcement, which may require further verification of the extent and nature of their involvement.

  2. Apr 2026
    1. Will smarter models be increasingly expensive because of greater accuracy or less expensive because they're smarter?

      作者提出一个非共识的二分法:大多数人认为AI模型要么因更精确而更贵,要么因更智能而更便宜。但作者暗示这两种趋势可能同时存在,形成锯齿状的成本模式,这挑战了人们对技术成本发展的线性预期。

    2. Then Opus 4.7 shipped & the smarter model became much more expensive. The cause : a new tokenizer

      大多数人认为AI模型变贵主要是因为能力提升,但作者揭示了一个反直觉的原因:更精确的分词器(tokenizer)导致需要处理更多token,从而使更智能的模型反而变得更贵。这挑战了'能力提升导致成本上升'的简单归因。

    3. When Anthropic launched Opus 4.5 in November 2025, the bigger, more expensive model was actually cheaper to use.

      大多数人认为更先进的AI模型必然更昂贵,但作者指出Claude Opus 4.5作为更大、更先进的模型实际上使用成本更低。这挑战了'先进=昂贵'的普遍认知,展示了AI效率提升可能带来的成本反直觉现象。

    1. The agent interprets new information and adapts the logic. The engine applies that logic continuously and emits precise updates.

      大多数人认为AI代理应该完全负责从数据收集到决策执行的整个流程。但作者提出颠覆性的观点:AI应该专注于逻辑解释和适应,而将执行和持续评估交给专门的数据库引擎。这种分工模式挑战了当前AI代理应该全能化的主流认知。

    2. Agents and CDC streams are powerful together because they split the work well.

      大多数人可能认为AI代理应该独立完成所有任务,包括数据获取和处理。但作者提出反直觉的分工模式:AI专注于逻辑解释和适应,而数据库引擎专注于持续评估和精确更新。这种分工挑战了当前AI代理应该端到端处理所有任务的主流观点。

    3. The fix is not smarter prompts. It is software built to meet agents halfway.

      大多数人认为提高AI性能的关键在于更好的提示工程或更智能的模型。但作者认为解决方案在于重新设计软件架构,使其与AI代理更好地协作,而不是继续改进AI本身。这是一个颠覆性的观点,挑战了当前AI开发的主流方向。

    4. Today's agents, the copilots, the chatbots are designed to be human like.

      大多数人认为AI助手应该模仿人类的交流方式,以便更好地与人类协作。但作者认为这种设计是错误的,因为它增加了认知负荷,违背了'平静技术'的理念。作者暗示AI应该更像是背景工具,而不是虚拟同事。

    5. The agent interprets new information and adapts the logic. The engine applies that logic continuously and emits precise updates.

      大多数人认为AI代理应该具备自主决策和执行能力。但作者提出了一种反直觉的分工模式:AI代理负责策略和逻辑调整,而执行引擎负责持续应用这些逻辑。这种模式将AI从'执行者'重新定位为'策略制定者',挑战了AI自主性的主流认知。

    6. Agents and CDC streams are powerful together because they split the work well.

      大多数人认为AI代理应该负责从端到端的任务执行。但作者认为AI代理和数据库引擎应该分工合作:代理负责解释新信息和调整逻辑,而数据库负责持续应用逻辑并发出精确更新。这种分工模式挑战了AI代理应该完全自主的主流观点。

    7. With change data capture (CDC), the system emits a stream of precise updates: inserts, updates, deletes, each tied to specific records.

      大多数人认为AI代理需要主动查询数据系统以获取信息。但作者提出了一种反直觉的方法:让数据库主动向AI代理发送变更事件,而不是让代理轮询或查询。这种模式将AI代理从主动查询者转变为被动响应者,从根本上改变了人机交互模式。

    8. The fix is not smarter prompts. It is software built to meet agents halfway.

      大多数人认为提高AI提示词质量是改善AI交互的关键。但作者认为真正解决方案是重新设计软件架构,使其与AI代理更好地协作,而不是改进提示词。这一观点颠覆了当前AI优化的主流方法,将焦点从AI本身转向系统设计。

    9. Today's agents, the copilots, the chatbots are designed to be human like.

      大多数人认为AI助手应该模仿人类交互方式,使其更自然、更易用。但作者认为这种设计方向是错误的,因为它需要高认知负荷来交互、解析和管理,违背了'平静技术'的理念。作者暗示我们应该让AI更像机器而非人类,以减少认知负担。

    1. Meanwhile, in reality, the only 'official' MeshCore is the github repo. It's the source of truth in terms of what is MeshCore, and Andy has never contributed to that.

      大多数人认为拥有商标或域名的人自然拥有项目的'官方'地位,但作者坚持只有GitHub仓库才是真正的'官方'来源,这挑战了知识产权与项目官方身份之间的常规认知。

    2. Since inception, the MeshCore development team have been working hard to build MeshCore. We've released more than 85 versions of the MeshCore Companion, Repeater and Room Server firmwares with support for more than 75 hardware variants. All of this has been hand crafted, by humans.

      在当今AI辅助编程盛行的时代,大多数人认为利用AI工具加速开发是理所当然的,但MeshCore团队坚持所有代码都是手工编写,这挑战了软件开发行业的效率优先共识。

    3. Andy Kirby did do an amazing job helping to promote the MeshCore project on his personal YouTube, but only promotes his own products now.

      大多数人认为项目贡献者应该持续推广整个项目生态系统,但作者暗示Andy从推广整个项目转向仅推广自己的产品,这种转变在开源社区中是罕见的,通常不被视为最佳实践。

    4. We have always been wary of AI generated code, but felt everyone is free to do what they want and experiment, etc.

      大多数人认为在软件开发中使用AI工具是提高效率和创新的合理方式,但作者团队明确表示他们一直对AI生成的代码持谨慎态度,这反映了在开源社区中对AI代码质量控制的非主流立场。

    1. LLM tend to use certain font combos like Space Grotesk, Instrument Serif and Geist

      大多数人认为AI能模仿任何设计风格,但作者指出AI实际上有特定的字体偏好,这揭示了AI设计的局限性而非无限可能性。这一发现挑战了我们对AI设计能力的认知,表明AI可能只是复制而非真正创新。

    2. I guess people will get back to crafting beautiful designs to stand out from the slop. On the other hand, I'm not sure how much design will still matter once AI agents are the primary users of the web.

      大多数人认为设计始终对用户体验至关重要,但作者质疑当AI成为主要网络用户时设计的重要性,这挑战了设计行业的核心假设。这一观点暗示设计可能从面向人类转向面向AI,彻底改变设计价值链。

    3. A designer recently told me that 'colored left borders are almost as reliable a sign of AI-generated design as em-dashes for text'

      大多数人认为AI设计难以识别,但作者认为简单的视觉元素如彩色边框就能可靠地识别AI生成的设计,这挑战了我们对AI设计复杂性的认知。这种观点暗示AI设计实际上有可预测的模式,而非完全无法捉摸。

    1. The good world is where everyone has AI, and not as a revokable privilege through an API, but through hard possession.

      大多数人可能认为通过API访问AI是民主化和可扩展的方式,但作者认为真正的AI民主化应该是通过硬所有权(hard possession),挑战了当前AI服务的主流商业模式。

    2. It works for Mars. I think there's so much value in colonizing Mars, and it's sad to me to see SpaceX diluting the mission buying up random AI bubble crap.

      大多数人可能认为AI和太空探索都是值得追求的目标,但作者认为这两者存在冲突,暗示SpaceX在AI领域的投资分散了其火星殖民的核心使命,挑战了科技多元化发展的共识。

    3. How does a normal person fit into Elon's world? What institutions will Elon leave behind? Is there any value in that society to art and culture?

      大多数人认为马斯克的愿景(如火星殖民)是积极和令人向往的,但作者质疑这种社会对普通人和文化艺术的价值,暗示马斯克的愿景可能创造一个缺乏人文关怀的社会。

    4. I can hear the rabid Elon fan defending him about Tesla patents or the Twitter algorithm or something, but those are not serious open source projects.

      大多数人认为埃隆·马斯克的开源贡献(如特斯拉专利)是值得称赞的,但作者认为这些并非真正的开源项目,暗示马斯克的开源承诺是表面性的,与真正的开源精神(如Linux和Kubernetes)有本质区别。

    5. Even the ideal version, industrial megaprojects at hyperhuman scale while constantly being out over your skis with leverage sounds hellish.

      大多数人认为大型AI项目和工业规模的发展是进步和繁荣的象征,但作者认为这种超人类规模的项目听起来像是地狱般的体验,因为它可能导致过度杠杆化和不可持续的压力。

    1. Commoditizing complements doesn't always work because focus is scarce even for the largest, fastest growing businesses.

      大多数人认为科技巨头拥有无限资源实施各种战略,但作者指出即使是最大、增长最快的企业也面临注意力稀缺问题。这一观点挑战了规模经济理论,暗示过度扩张可能导致核心竞争力的稀释。

    2. Some categories never developed a competitive response to this strategy : email, advertising infrastructure, user-generated video.

      大多数人认为所有商业领域都有能力应对颠覆性竞争,但作者指出某些类别如电子邮件、广告基础设施等从未找到有效的竞争对策。这暗示了某些市场结构可能存在根本性弱点,无法通过传统竞争策略应对免费化浪潮。

    3. The commoditization flywheel : both companies give away complements to drive usage of the core.

      大多数人认为AI公司应该专注于核心产品并保持其专有性,但作者认为AI巨头应该效仿谷歌,通过免费提供互补产品来推动核心产品的使用,这与传统科技公司的护城河策略相悖。

    4. Commoditizing complements doesn't always work because focus is scarce even for the largest, fastest growing businesses.

      大多数人认为科技巨头拥有无限资源可以实施任何战略,但作者指出即使是最大的企业也面临注意力稀缺问题。这与对科技巨头的普遍认知相悖,暗示规模优势也有其局限性。

    5. Some categories never developed a competitive response to this strategy : email, advertising infrastructure, user-generated video.

      大多数人认为市场竞争最终会形成平衡,所有行业都会有相应的竞争策略,但作者指出有些类别从未对免费化策略形成有效回应,这表明市场并非总是自我调节,存在结构性失衡。这一观点挑战了自由市场的完美竞争理论。

    6. The risk of this strategy to the ecosystem is that it makes previously attractive categories no longer viable. Commoditizing the complement does not demand a best-in-class replacement.

      大多数人认为市场竞争总是促进产品质量提升和创新,但作者认为谷歌和Anthropic的免费化策略实际上可能扼杀某些行业的创新动力,因为'足够好'的免费产品就足以改变市场动态,这与传统经济学中的竞争理论相悖。

    1. Several correlated but not strictly identical changes happened over the same few months: scaling inference compute, heavier use of RL in post-training, and models producing reasoning tokens.

      大多数人可能将AI能力加速归因于单一因素(如模型规模增大),但作者指出这是多种因素共同作用的结果,包括推理计算扩展、强化学习在训练后阶段的使用增加以及模型生成推理标记的能力。这一多元归因挑战了单一因素决定论。

    2. Tasks where correctness is harder to verify may not have seen the same speedup, so the acceleration we document here may not be as general as the headline numbers suggest.

      大多数人可能被媒体报道的AI加速数据所影响,认为所有AI任务都在加速,但作者明确指出,那些正确性难以验证的任务可能没有相同的加速速度。这一观点挑战了人们对AI能力普遍加速的乐观预期。

    3. The three metrics where we find acceleration are concentrated in programming and mathematics. These are areas that labs have explicitly targeted for improvement, and they share an important property: correctness is easy to verify automatically.

      大多数人可能认为AI能力的加速是跨领域普遍发生的,但作者指出加速主要集中在编程和数学领域,因为这些领域正确性容易自动验证。这一发现挑战了人们对AI能力普遍提升的假设,暗示加速可能是有选择性的。

    4. Our fourth metric, an index constructed from WeirdML V2 results, showed no sign of acceleration. A single global linear trend fit the data best.

      大多数人可能认为所有AI能力指标都应该同步加速,但作者发现WeirdML V2指标没有显示出任何加速迹象,最佳拟合仍是简单的全局线性趋势。这一发现表明AI能力的加速并不是普遍现象,而是特定于某些任务领域。

    5. Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

      大多数人认为不同AI模型之间的性能差异是渐进式的,但作者发现推理模型不仅一次性实现了性能跃升,而且以比非推理模型快2-3倍的速度持续进步。这一发现挑战了人们对AI模型性能提升方式的常规理解。

    6. Three of the four metrics (ECI, log METR 50% time horizon, and a math-focused index we constructed from several math benchmarks) show strong evidence that progress has sped up relative to a global linear trend fit to data from 2023 onward.

      大多数人认为AI能力提升是渐进式的线性发展,但作者通过数据分析发现,在三个关键指标上,AI能力实际上已经加速,这挑战了人们对AI发展速度的普遍认知。这种加速现象发生在2023年之后,与推理模型的发布时间点吻合。

    7. Several correlated but not strictly identical changes happened over the same few months: scaling inference compute, heavier use of RL in post-training, and models producing reasoning tokens.

      大多数人可能将AI进步归因于单一因素(如模型规模或数据量),但作者指出推理能力的提升是多种因素共同作用的结果,包括推理计算扩展、强化学习更广泛应用以及模型产生推理标记等。这挑战了人们对AI进步驱动因素的认知。

    8. Tasks where correctness is harder to verify may not have seen the same speedup, so the acceleration we document here may not be as general as the headline numbers suggest.

      主流媒体和公众可能认为AI能力在所有领域都在加速提升,但作者明确指出,在正确性难以验证的任务中可能没有相同的加速现象。这一观点挑战了人们对AI进步普遍性的假设。

    9. WeirdML V2 places models in an unusually resource-constrained environment: models get only five attempts to submit working code, with no access to external tools. This setup has not been the focus of recent RL training.

      大多数人可能认为所有AI评估指标都会反映相同的进步趋势,但研究发现WeirdML V2指标没有显示加速,因为它设置了资源限制环境,而近期强化学习训练并未关注此类设置。这表明AI进步可能受评估方法的影响。

    10. The three metrics where we find acceleration are concentrated in programming and mathematics. These are areas that labs have explicitly targeted for improvement, and they share an important property: correctness is easy to verify automatically.

      主流观点可能认为AI能力在各个领域的提升是均衡的,但作者指出加速现象主要集中在编程和数学领域,因为这些领域的正确性容易自动验证。这暗示AI进步可能不是普遍性的,而是集中在特定可量化的领域。

    11. Reasoning models show both a one-off jump in performance and a roughly 2-3x faster trend compared to non-reasoning models.

      大多数人可能认为不同类型的AI模型性能提升速度大致相同,但研究发现推理模型不仅有一次性的性能飞跃,而且提升速度是非推理模型的2-3倍。这一发现颠覆了人们对不同模型类型进步速度的预期。

    12. Three of four metrics show strong evidence of acceleration, seemingly driven by reasoning models.

      大多数人认为AI能力提升是渐进式的线性增长,但作者通过数据分析发现,在四个关键能力指标中有三个出现了明显加速,且这种加速似乎与推理模型的出现直接相关。这挑战了人们对AI进步速度的普遍认知。

    13. Three of four metrics show strong evidence of acceleration, seemingly driven by reasoning models.

      大多数人认为AI能力的发展是持续稳定的线性增长,但作者通过数据分析发现,在四个关键指标中有三个显示出明显的加速趋势,这种加速是由推理模型驱动的。这一结论挑战了人们对AI进步速度的常规认知,表明2024年推理模型的引入可能标志着AI能力发展模式的转变。

    1. Within eight days, the same campaign had cascaded from GitHub Actions to Docker Hub, npm, PyPI, and the VS Code extension marketplace. With just one token across five ecosystems, thousands of organizations were potentially impacted.

      大多数人认为软件供应链攻击通常是针对特定生态系统或缓慢扩散的,但作者展示了跨生态系统的快速级联攻击。这种攻击速度和范围远超传统认知,表明现代软件供应链的脆弱性被严重低估。

    2. Modern-day security tooling looks for the wrong things. Most software composition analysis tools work by checking your dependencies against a database of known vulnerabilities – CVEs. But a deliberately planted backdoor doesn't have a CVE.

      大多数安全团队依赖CVE数据库来评估风险,但作者指出这种方法对故意植入的后门完全无效。这一观点挑战了行业共识,暗示现有安全工具在新型供应链攻击面前已经过时,需要转向行为分析等新方法。

    3. The result is a mismatch that should terrify anyone building software: the attack surface is expanding faster than any human can monitor, and the entities making dependency decisions are increasingly not human.

      大多数人认为安全问题可以通过增加人力监控和审查来解决,但作者认为在AI时代,攻击面扩展速度已经超过了人类监控能力,且依赖决策越来越由AI而非人类做出。这一观点挑战了传统安全理念,暗示需要全新的自动化防御机制。

    1. Liam Price just cracked a 60-year-old problem that world-class mathematicians have tried and failed to solve.

      大多数人认为解决长期未解的数学难题需要顶尖数学家的专业知识和多年研究,但作者认为一个业余爱好者通过AI就做到了,这挑战了数学专业壁垒的传统观念。

    2. An AI researcher subsequently gifted them each a ChatGPT Pro subscription to encourage their 'vibe mathing.'

      大多数人认为严肃的数学研究需要严谨的方法和深厚的专业知识,但作者使用'vibe mathing'这种非正式术语描述这种研究方式,挑战了学术研究方法论的传统规范。

    3. We have discovered a new way to think about large numbers and their anatomy. It's a nice achievement. I think the jury is still out on the long-term significance.

      大多数人认为AI的数学突破具有重大意义,但作者认为其长期意义尚不确定,这挑战了人们对AI数学成就重要性的普遍预期,暗示技术突破不一定等同于长期价值。

    4. The LLM took an entirely different route, using a formula that was well known in related parts of math, but which no one had thought to apply to this type of question.

      大多数人认为数学突破需要全新的理论或方法,但作者认为AI只是将已知公式应用到新领域就能取得突破,这挑战了人们对数学创新本质的理解,暗示创新有时来自于跨领域应用而非全新创造。

    5. Liam Price just cracked a 60-year-old problem that world-class mathematicians have tried and failed to solve. He's 23 years old and has no advanced mathematics training.

      大多数人认为解决重大数学问题需要深厚的专业训练和多年经验,但作者认为一个23岁没有高级数学训练的业余人士也能解决60年悬而未决的问题,这挑战了学术界对专业资质的传统认知。

    6. Liam Price just cracked a 60-year-old problem that world-class mathematicians have tried and failed to solve. He's 23 years old and has no advanced mathematics training.

      大多数人认为解决复杂的数学问题需要深厚的专业训练和多年经验,但作者认为一个没有高级数学训练的23岁年轻人仅凭AI工具就能解决困扰顶级数学家60年的问题,这挑战了数学领域的专业壁垒认知。

    7. What he does have is a ChatGPT Pro subscription, which gives him access to the latest large language models from OpenAI.

      大多数人认为数学成就主要依赖于个人智力和训练,但Price的成功关键是他拥有AI工具访问权限,这暗示在未来的数学领域,技术资源可能比个人能力更重要,挑战了传统天才观念。

    8. Lichtman tried to prove this, too, but got stuck like everyone else before him.

      大多数人认为数学突破来自于持续不断的努力和渐进式改进,但Lichtman和其他专家的失败表明,有时问题不在于努力程度而在于思维方式的局限,这挑战了我们对数学进步过程的认知。

    9. An AI researcher subsequently gifted them each a ChatGPT Pro subscription to encourage their 'vibe mathing.'

      大多数人认为严肃的数学研究需要严谨的方法和深厚的理论基础,但研究人员用'vibe mathing'这种非正式方式描述他们的工作,暗示数学发现可能源于看似随性的探索而非严格的规划。

    10. I had the intuition that these problems were kind of clustered together and they had some kind of unifying feel to them. And this new method is really confirming that intuition.

      大多数人认为数学问题是孤立的,需要不同的方法解决,但Lichtman的直觉表明这些问题可能有内在联系,AI的发现证实了这一观点,暗示数学领域可能存在尚未被发现的深层统一性。

    11. The LLM took an entirely different route, using a formula that was well known in related parts of math, but which no one had thought to apply to this type of question.

      大多数人认为数学突破需要全新的理论或方法,但AI的解决方案使用了已知公式只是应用到了新领域,这表明创新可能更多来自于跨领域应用而非全新发明,挑战了我们对数学创新本质的理解。

    12. Liam Price just cracked a 60-year-old problem that world-class mathematicians have tried and failed to solve. He's 23 years old and has no advanced mathematics training.

      大多数人认为解决复杂的数学难题需要深厚的专业训练和多年经验,但这个案例表明,一个没有高级数学训练的23岁年轻人仅通过AI工具就解决了困扰顶尖数学家60年的问题,挑战了专业知识在数学突破中的必要性。

    13. I had the intuition that these problems were kind of clustered together and they had some kind of unifying feel to them. And this new method is really confirming that intuition.

      大多数人认为数学问题是孤立且独特的,每个问题需要专门的解决方法,但作者认为AI的发现证实了数学问题之间存在某种统一性和关联性,这挑战了人们对数学问题独立性的传统认知。

    14. The LLM took an entirely different route, using a formula that was well known in related parts of math, but which no one had thought to apply to this type of question.

      大多数人认为数学突破需要全新的理论和创新方法,但作者认为AI能够通过重新组合和应用现有知识来解决问题,这挑战了人们对创新必须来自全新理论的认知,展示了AI独特的知识连接能力。

    1. This card was updated on April 24, 2026, to include additional information about safeguards for the deployment of GPT‑5.5 and GPT‑5.5 Pro in the API.

      大多数人认为系统卡应该在发布时包含所有相关信息,不需要后续更新,但OpenAI在发布后仅一天就更新了系统卡以增加API部署的安全措施信息。这挑战了科技产品文档管理的常规做法,暗示AI安全措施是动态发展的,需要持续调整,这违背了传统软件发布中'文档一次性完成'的共识。

    2. We separately evaluate GPT‑5.5 Pro in certain cases because we judge that the setting could materially impact the relevant risks or appropriate safeguards posture.

      大多数人认为如果两个模型使用相同的基础架构,它们的风险和安全需求应该相似,但OpenAI明确表示GPT-5.5 Pro需要单独评估,因为'设置可能显著影响相关风险或适当的安全措施立场'。这挑战了AI评估领域普遍认为的'相同基础模型的安全特性一致'的共识,暗示即使是微小的设置变化也可能导致显著不同的风险特征。

    3. We are releasing GPT‑5.5 with our strongest set of safeguards to date, designed to reduce misuse while preserving legitimate, beneficial uses of advanced capabilities.

      大多数人认为更强的安全限制会不可避免地限制AI的功能和实用性,但OpenAI声称他们能够同时实现'减少滥用'和'保留合法、有益的高级功能使用'。这挑战了AI安全领域普遍存在的'安全与功能之间存在权衡'的共识,暗示他们已经找到了一种创新的方法,可以在不牺牲功能的情况下增强安全性。

    4. GPT‑5.5 understands the task earlier, asks for less guidance, uses tools more effectively, checks it work and keeps going until it's done.

      大多数人认为AI模型需要持续的人工指导和监督才能完成复杂任务,但作者声称GPT-5.5能够'理解任务更早,要求更少的指导,更有效地使用工具,检查工作并持续进行直到完成'。这挑战了AI领域普遍认为的'当前AI系统仍需大量人类监督'的共识,暗示GPT-5.5已经实现了更高程度的自主性。

    5. We subjected the model to our full suite of predeployment safety evaluations and our Preparedness Framework, including targeted red-teaming for advanced cybersecurity and biology capabilities

      大多数人认为AI安全评估主要集中在防止直接有害输出,但OpenAI特别强调了对'高级网络生物学能力'的针对性红队测试。这暗示GPT-5.5可能具有比预期更强大的生物相关能力,这违背了AI领域普遍认为的'语言模型主要处理文本信息'的共识,表明AI已经深入到专业科学领域。

    1. Testing universal jailbreaks for biorisks in GPT‑5.5

      大多数人认为AI安全测试应专注于防止有害内容生成,但OpenAI主动邀请研究人员寻找'通用越狱方法'来突破生物安全限制,这挑战了传统安全思维,表明他们认为主动寻找漏洞比被动防御更有效。

    1. We believe this is what drove the separate reports of usage limits draining faster than expected.

      大多数人会直接将API使用量异常归因于用户行为或模型本身,但作者揭示了一个技术实现细节(缓存bug)如何间接导致使用量异常。这挑战了常规问题归因逻辑,展示了系统组件间的意外互动如何产生看似无关的问题表象。

    2. As part of this investigation, we ran more ablations (removing lines from the system prompt to understand the impact of each line) using a broader set of evaluations. One of these evaluations showed a 3% drop for both Opus 4.6 and 4.7.

      大多数人认为微小的系统提示变更只会带来微不足道的影响,但作者展示了一个看似微不足道的提示变更(限制字数)却导致了3%的性能下降。这挑战了'小变更小影响'的直觉认知,揭示了AI系统中微小变化可能带来的非线性影响。

    3. After multiple weeks of internal testing and no regressions in the set of evaluations we ran, we felt confident about the change and shipped it alongside Opus 4.7 on April 16.

      大多数人认为充分的内部测试可以预防产品发布后的重大问题,但作者展示了一个经过数周内部测试且没有发现问题的系统提示变更却导致了明显的质量下降。这挑战了'测试覆盖率等于产品质量'的传统观念,暗示了评估指标与实际用户体验之间可能存在巨大鸿沟。

    4. Two unrelated experiments made it challenging for us to reproduce the issue at first: an internal-only server-side experiment related to message queuing; and an orthogonal change in how we display thinking suppressed this bug in most CLI sessions

      大多数人认为复杂的系统测试流程应该能够发现大多数关键缺陷,但作者展示了即使有多重测试机制,两个看似无关的实验如何协同掩盖了一个严重bug。这挑战了'全面测试能保证产品质量'的传统认知,揭示了系统复杂性带来的意外风险。

    5. In our internal evals and testing, medium effort achieved slightly lower intelligence with significantly less latency for the majority of tasks.

      大多数人认为内部评估和测试足以代表用户真实体验,但作者承认他们的内部测试未能准确捕捉到用户对AI智能度的实际感知差异。这暗示了实验室环境与实际使用场景之间存在根本性脱节,挑战了传统产品测试方法论的有效性。

    6. We reverted this change on April 7 after users told us they'd prefer to default to higher intelligence and opt into lower effort for simple tasks.

      大多数人认为AI系统应该优化速度和效率,但作者认为用户更愿意默认选择更高智能而非更低延迟,这挑战了产品优化的常规思维。用户宁愿忍受偶尔的延迟也要换取更高的代码质量,这违背了大多数科技公司追求'更快更省'的常规做法。

    1. The products will need to get worse, more expensive, or both if VCs are to get their money back.

      主流观点认为科技公司会通过产品创新和改进来提高价值,但作者直言AI公司可能需要让产品变得更差或更昂贵才能满足投资者回报要求,这挑战了科技行业进步的叙事,揭示了资本压力与产品价值之间的潜在冲突。

    1. the system achieved this training result more than 20 times faster than conventional synchronization methods.

      大多数人认为分布式训练由于需要同步和通信,必然比单机训练慢,但作者认为Decoupled DiLoCo比传统同步方法快20倍以上,这挑战了人们对分布式训练速度的固有认知,展示了异步计算的潜力。

    2. chips from different generations running at different speeds still matched the ML performance of single-chip-type training runs, ensuring that even older hardware can meaningfully accelerate AI training.

      大多数人认为混合不同代际的硬件进行训练会降低性能或效率,但作者认为即使不同代际、不同速度的芯片混合使用,仍能达到与单一芯片类型训练相同的机器学习性能,这挑战了硬件必须同质化的行业共识。

    3. With increasing levels of hardware failure, Decoupled DiLoCo continues to deliver a high level of 'goodput', or useful training, while that of other approaches nosedives.

      大多数人认为硬件故障会显著降低分布式训练的效率和性能,但作者认为即使在硬件故障率极高的环境下,Decoupled DiLoCo仍能保持88%的有效训练率,而传统方法则暴跌至27%,这挑战了人们对故障容忍能力的传统认知。

    4. By dividing large training runs across decoupled 'islands' of compute, with asynchronous data flowing between them, this architecture isolates local disruptions so that other parts of the system can keep learning efficiently.

      大多数人认为分布式AI训练需要高度同步和紧密耦合的系统才能保证效率,但作者认为通过解耦的'计算岛屿'架构,即使局部硬件故障,系统其他部分仍能高效学习,因为故障被隔离了。这挑战了传统分布式训练必须保持同步的主流认知。

    1. Amazon is investing $5 billion in Anthropic today, with up to an additional $20 billion in the future. This builds on the $8 billion Amazon has previously invested.

      大多数人认为科技巨头对AI公司的投资通常在数亿级别,但Amazon对Anthropic的总投资可能高达330亿美元,这远超行业共识。这种规模的投资表明科技巨头对AI基础设施的重视程度和投入规模正在以前所未有的方式增长,可能重塑AI行业的资本结构和竞争动态。

    2. Claude remains the only frontier AI model available to customers on all three of the world's largest cloud platforms: AWS (Bedrock), Google Cloud (Vertex AI), and Microsoft Azure (Foundry).

      大多数人认为AI模型通常会与单一云平台深度绑定,形成生态系统锁定,但Claude同时出现在三大云平台上,这挑战了AI行业平台绑定策略的主流认知。这种多平台策略可能预示着AI模型提供商正寻求更大的市场覆盖和避免单一平台依赖,改变行业竞争格局。

    3. Anthropic will also use incremental capacity for Claude in Amazon Bedrock. The agreement includes expansion of inference in Asia and Europe to better serve Claude's growing international customer base.

      大多数人认为AI模型主要在美国市场发展,但Anthropic明确表示正在大力扩展亚洲和欧洲市场,这挑战了AI服务主要集中在美国的共识。这种全球扩张速度表明AI市场的地理分布正在迅速多元化,可能重塑全球AI产业格局。

    4. Our run-rate revenue has now surpassed $30 billion, up from approximately $9 billion at the end of 2025.

      大多数人认为AI公司仍处于烧钱阶段,难以实现盈利,但Anthropic的收入在短短几个月内增长了三倍多,达到300亿美元的年化收入。这一惊人的增长速度挑战了AI行业普遍亏损的共识,表明AI模型商业化可能比预期更快、规模更大。

    5. We have signed a new agreement with Amazon that will deepen our existing partnership and secure up to 5 gigawatts (GW) of capacity for training and deploying Claude

      大多数人认为AI公司主要依赖通用GPU芯片训练模型,但Anthropic与Amazon的合作表明他们正大规模采用专用AI芯片(Trainium),这挑战了行业对通用芯片依赖的主流认知。5GW的容量远超大多数AI公司的规模,反映了专用芯片在AI训练中的经济性和效率优势正在被重新评估。

    1. The Prompt API uses the Gemini Nano model in Chrome. While the API is built into Chrome, the model is downloaded separately the first time an origin uses the API.

      大多数人认为内置API应该包含所有必要组件,无需额外下载,但作者明确指出模型需要单独下载。这与人们对'内置'API应该即开即用的普遍认知相悖,暗示用户首次使用时可能会面临显著的下载时间和存储压力。

    2. The Prompt API for the web is still being developed. While we build this API, refer to our best practices on session management for optimal performance.

      大多数人认为浏览器AI功能应该是成熟且生产就绪的,但作者明确表示该API仍在开发中。这与人们对Chrome作为成熟浏览器应该提供稳定可靠功能的认知相悖,暗示AI功能可能还不够稳定,需要开发者额外注意性能优化。

    3. The network requirement is only for the initial download of the model. Subsequent use of the model does not require a network connection. No data is sent to Google or any third party when using the model.

      大多数人认为使用Google的AI模型必然会涉及数据传输和隐私问题,但作者强调模型完全在设备上运行且不向Google发送数据。这与人们对大型科技公司AI服务通常涉及数据收集的普遍认知相悖,暗示Chrome的AI功能可能比想象的更加注重隐私保护。

    4. The Prompt API isn't available in Web Workers for now, due to the complexity of establishing a responsible document for each worker in order to check the permissions policy status.

      大多数人认为现代浏览器API应该支持Web Workers以实现并行处理,但作者明确表示Prompt API不支持Web Workers。这与人们对浏览器API应该全面支持现代Web开发模式的认知相悖,限制了开发者在后台线程中使用AI的能力。

    1. Microsoft continues to participate directly in OpenAI's growth as a major shareholder.

      大多数人认为在修改了合作协议后,微软可能会减少其在OpenAI的股权投资,但作者认为微软仍然是OpenAI的主要股东,这表明尽管合作关系有所调整,但双方仍然保持着深度的利益绑定,这可能是一种非传统的长期战略伙伴关系模式。

    2. Revenue share payments from OpenAI to Microsoft continue through 2030, independent of OpenAI's technology progress, at the same percentage but subject to a total cap.

      大多数人认为随着OpenAI技术的发展,其对微软的支付可能会增加或调整,但作者认为OpenAI对微软的支付将保持固定比例且有上限,这表明OpenAI正在寻求更可预测的财务安排,不受技术进步的影响,这可能是一种反直觉的风险管理策略。

    3. Microsoft will continue to have a license to OpenAI IP for models and products through 2032. Microsoft's license will now be non-exclusive.

      大多数人认为微软会寻求对OpenAI技术的独家使用权,以保持其在AI领域的竞争优势,但作者认为微软的许可权变为非独家,这打破了传统科技合作中的排他性模式,暗示OpenAI正在向更开放的合作方式转变,可能为其他合作伙伴铺平道路。

    4. Microsoft will no longer pay a revenue share to OpenAI.

      大多数人认为微软作为OpenAI的主要投资者和合作伙伴,会继续通过收入分成来支持OpenAI的发展,但作者认为微软已经改变了这一模式,这可能表明微软认为OpenAI的技术已经足够成熟,不再需要这种财务激励,或者微软有其他方式从合作中获益。

    5. OpenAI can now serve all its products to customers across any cloud provider.

      大多数人认为OpenAI会完全依赖微软Azure云服务,因为微软是其主要投资者和合作伙伴,但作者认为OpenAI现在拥有了多云策略的灵活性,这打破了科技巨头间典型的排他性合作模式,暗示OpenAI正在寻求更大的自主权和市场机会。

    1. The compliance-driven buyers improvising local AI out of retail Mac Minis because the product they need does not exist.

      大多数人认为企业AI采用需要专门的解决方案和供应商,但作者指出一些合规驱动的买家正在使用零售版Mac Mini自行构建本地AI解决方案。这挑战了企业AI市场的传统认知,暗示市场可能存在未被满足的需求,以及企业正在以非传统方式应对AI挑战。

    2. Why the company that moved computing off the mainframe fifty years ago is making the same structural move with AI, and what that predicts.

      大多数人将苹果的AI战略视为孤立的商业决策,但作者将其与苹果历史上将计算从大型机转移到个人电脑的战略相提并论。这提供了一个反直觉的历史视角,暗示苹果可能正在引领AI从集中式云服务向分布式设备端的范式转变,挑战了当前AI行业向云端集中化的主流趋势。

    3. The question it forces is not which model is best. It is who owns the inference layer your organization depends on, what happens when the economics of that layer stop being subsidized, and whether the thing in your pocket turns out to matter more than the thing in the datacenter.

      大多数人关注AI模型本身的性能和优势,但作者认为真正关键的是谁拥有推理层以及其经济可持续性。这挑战了当前AI行业的主流关注点,暗示未来竞争的核心将从模型本身转向推理层的控制和成本结构,这是一个反直觉的视角转换。

    4. The structural cost problem in AI inference that makes Apple's on-device bet defensible, not just defensive.

      大多数人认为苹果转向设备端AI只是防御性策略,因为他们在云AI领域落后,但作者认为这是基于对AI推理层经济结构问题的深刻理解而做出的主动选择。这挑战了主流对苹果AI战略的看法,暗示设备端AI可能比我们想象的更具经济优势。

    5. The board looked at the AI race Apple was losing and, rather than try harder at the thing that was failing, changed which game the company plays.

      大多数人认为面对竞争失败,公司应该加倍投入资源在原有领域追赶,但作者认为苹果选择了完全不同的策略——改变游戏规则而非在原有规则下竞争。这挑战了传统商业战略思维,暗示苹果可能正在从云AI转向设备端AI,这是一种颠覆性的战略转向。

    6. For a company that spent fifteen years running a functional model where no single discipline owned a product, putting two hardware engineers at the top is not a personnel decision. It is a structural break.

      大多数人认为苹果的高层变动只是常规的人事调整,但作者认为这是苹果在AI竞争中失败后采取的结构性变革,反映了公司战略的根本转变。这挑战了我们对科技公司领导层变动的常规认知,暗示苹果正在从功能型组织转向以硬件为中心的结构,以应对AI挑战。

    1. This means that improvements on SWE-bench Verified no longer reflect meaningful improvements in models' real-world software development abilities. Instead, they increasingly reflect how much the model was exposed to the benchmark at training time.

      大多数人认为基准测试分数的提高意味着模型实际能力的提升。但作者明确表示,SWE-bench Verified的改进不再反映模型真实软件开发能力的进步,而是更多地反映了模型在训练时接触该基准测试的程度。这一结论挑战了整个AI评估体系的有效性,暗示我们可能需要重新思考如何衡量AI的真实进步。

    2. Tests reject correct solutions: We audited a 27.6% subset of the dataset that models often failed to solve and found that at least 59.4% of the audited problems have flawed test cases that reject functionally correct submissions

      大多数人认为代码测试是客观公正的,能够准确评估模型的真实能力。但作者发现,近60%的测试案例存在缺陷,会拒绝功能上正确的解决方案。这一发现挑战了AI评估领域的共识,表明我们广泛使用的基准测试可能存在系统性问题,无法准确反映模型的实际编程能力。

    1. Our RL infra team used a K2.6-backed agent that operated autonomously for 5 days, managing monitoring, incident response, and system operations, demonstrating persistent context, multi-threaded task handling, and full-cycle execution from alert to resolution.

      大多数人认为AI代理系统难以长时间持续运行,通常会面临注意力分散、上下文丢失或性能下降的问题。但作者展示的AI系统能够连续5天自主管理复杂的技术运维工作,这挑战了人们对AI代理持续运行能力的传统认知,暗示AI可能已经具备接近人类的持久工作能力。

    2. Kimi K2.6 autonomously overhauled exchange-core, an 8-year-old open-source financial matching engine. Over a 13-hour execution, the model iterated through 12 optimization strategies, initiating over 1,000 tool calls to precisely modify more than 4,000 lines of code.

      大多数人认为AI在复杂工程任务中仍需要人类专家的指导和监督,难以独立完成大规模系统重构。但作者展示了AI能够自主分析、优化并重构一个运行8年的金融系统,这挑战了人们对AI工程能力的传统认知,暗示AI可能已经具备系统级架构设计和优化的能力。

    1. NEC will establish a Center of Excellence to develop a highly skilled, AI-enabled engineering organization

      大多数人认为AI会使专业知识和技能贬值,但作者认为AI实际上需要更高水平的工程专业知识,因为企业正在建立专门的卓越中心来培养AI技能,这表明AI工具正在提升而非降低工程工作的专业门槛。

    2. As part of its long-running Client Zero initiative, in which NEC serves as its own first customer before offering its technology to clients

      大多数人认为企业会先开发产品然后内部使用,但作者认为NEC采用了反向策略,先内部大规模应用AI技术然后再向客户推广,这表明企业正在采用更激进的方法来验证和改进AI解决方案,挑战了传统的产品开发流程。

    3. NEC aims to build one of Japan's largest AI-native engineering teams, who will use Claude Code in their work.

      大多数人认为AI会取代大量工程师职位,但作者认为AI实际上是在创造新的工程角色和技能需求,因为NEC正在积极建立一支大规模的AI原生工程团队,这表明AI工具正在增强而非替代工程能力,创造新的就业机会。

    1. Claude packages everything into a handoff bundle that you can pass to Claude Code with a single instruction.

      大多数人认为设计和开发是两个分离的专业领域,需要专门的交接流程和工具,但作者暗示AI可以实现从设计到开发的无缝单指令转换。这一观点挑战了软件开发与设计之间的传统界限,暗示AI可能重新定义跨职能协作的方式。

    2. Our most complex pages, which took 20+ prompts to recreate in other tools, only required 2 prompts in Claude Design.

      大多数人认为复杂的设计任务需要更多的提示和人工干预,但作者声称他们的AI工具能用更少的提示完成更复杂的设计。这一观点挑战了人们对AI设计工具复杂度与输入量关系的普遍认知,暗示AI可能在某些方面比人类更擅长处理复杂性。

    3. What used to take a week of back-and-forth between briefs, mockups, and review rounds now happens in a single conversation.

      大多数人认为设计过程必然需要多轮迭代和长时间沟通,但作者声称AI可以将这一过程缩短到单次对话完成。这一观点挑战了设计工作流程的传统认知,暗示AI可能彻底改变设计协作的时间框架和效率预期。

    4. Claude Design gives designers room to explore widely and everyone else a way to produce visual work.

      大多数人认为设计专业技能是创造高质量视觉作品的必要条件,但作者认为AI工具可以让非专业人士也能生产专业水平的视觉作品。这一观点挑战了设计专业性的传统观念,暗示专业技能可能不再是高质量设计的唯一门槛。

    5. Even experienced designers have to ration exploration—there's rarely time to prototype a dozen directions, so you limit yourself to a few.

      大多数人认为专业设计师拥有充分的创意自由和资源来探索多种设计方案,但作者认为即使是经验丰富的设计师也受到时间和资源的严重限制,只能探索少数几个方向。这一观点挑战了人们对设计行业创意过程的普遍认知,揭示了设计实践中的现实约束。

    1. The gains are especially strong in agentic coding, computer use, knowledge work, and early scientific research—areas where progress depends on reasoning across context and taking action over time.

      大多数人认为AI进步主要体现在特定领域的知识获取和模式识别上,而非跨上下文的推理和长期行动能力。但作者强调GPT-5.5在需要持续推理和行动的领域取得显著进步,这一观点挑战了AI能力发展的主流叙事,暗示通用智能可能比预期更早实现。

    2. GPT‑5.5 found a proof of a longstanding asymptotic fact about off-diagonal Ramsey numbers, later verified in Lean. The result is a concrete example of GPT‑5.5 contributing not just code or explanation, but a surprising and useful mathematical argument in a core research area.

      大多数人认为AI在数学研究领域仅能辅助计算或提供解释,无法独立进行创造性数学推理。但作者展示GPT-5.5能够发现并证明数学定理,这一突破挑战了数学研究作为纯粹人类活动的传统观念,暗示AI可能成为真正的'研究伙伴'而非仅是工具。

    3. We are treating the biological/chemical and cybersecurity capabilities of GPT‑5.5 as High under our Preparedness Framework. While GPT‑5.5 didn't reach Critical cybersecurity capability level, our evaluations and testing showed that its cybersecurity capabilities are a step up compared to GPT‑5.4.

      大多数人认为AI在网络安全领域的应用主要局限于防御辅助,而非直接参与核心安全任务。但作者暗示GPT-5.5已具备'高级'网络安全能力,这一分类表明AI已从被动防御工具向主动安全参与者转变,挑战了网络安全领域对人类主导地位的认知。

    4. Losing access to GPT‑5.5 feels like I've had a limb amputated.

      大多数人将AI工具视为辅助性资源,失去后只会带来不便而非功能丧失。但这位NVIDIA工程师的比喻表明,GPT-5.5已从辅助工具转变为不可或缺的'认知延伸',这种依赖程度远超当前主流认知中人与AI的关系定位,暗示了人机协作范式的根本性转变。

    5. GPT‑5.5 delivers this step up in intelligence without compromising on speed: larger, more capable models are often slower to serve, but GPT‑5.5 matches GPT‑5.4 per-token latency in real-world serving, while performing at a much higher level of intelligence.

      大多数人认为更强大的AI模型必然伴随着更高的计算成本和更慢的响应速度,但作者认为GPT-5.5打破了这一规律,实现了更高的智能水平与相同的延迟时间并存。这一反直觉的发现挑战了AI领域'能力与效率成反比'的传统认知,暗示模型架构优化可能比单纯扩大规模更有效。

    6. GPT‑5.5 found a proof of a longstanding asymptotic fact about off-diagonal Ramsey numbers, later verified in Lean. The result is a concrete example of GPT‑5.5 contributing not just code or explanation, but a surprising and useful mathematical argument in a core research area.

      大多数人认为AI在数学研究中的作用主要是辅助计算和验证,但作者认为GPT-5.5能够独立发现数学证明,这在数学研究领域是革命性的。这一观点挑战了人们对AI在创造性思维和抽象推理领域能力的传统认知,暗示AI可能正在从工具转变为研究伙伴。

    7. The viable path is trusted access, robust safeguards that scale with capability, and the operational capacity to detect and respond to serious misuse.

      大多数人认为AI安全应该通过限制访问和严格监管来实现,但作者认为'可信访问'结合'随能力扩展的保障措施'才是可行路径。这一观点挑战了传统的AI安全治理理念,暗示过度限制可能会阻碍AI防御能力的充分发挥,而平衡的开放与安全才是最佳策略。

    8. We are treating the biological/chemical and cybersecurity capabilities of GPT‑5.5 as High under our Preparedness Framework. While GPT‑5.5 didn't reach Critical cybersecurity capability level, our evaluations and testing showed that its cybersecurity capabilities are a step up compared to GPT‑5.4.

      大多数人认为AI在网络安全领域的进步应该是渐进式的,但作者暗示GPT-5.5代表了网络安全能力的显著跃升,达到了'高'级别而非仅仅'临界'级别。这一观点挑战了人们对AI安全能力发展速度的预期,暗示AI在防御复杂网络威胁方面可能比人们想象的进步更快。

    9. GPT‑5.5 delivers this step up in intelligence without compromising on speed: larger, more capable models are often slower to serve, but GPT‑5.5 matches GPT‑5.4 per-token latency in real-world serving, while performing at a much higher level of intelligence.

      大多数人认为更强大的AI模型必然伴随着更高的计算成本和更慢的响应速度,但作者认为GPT-5.5打破了这一权衡关系,实现了更高智能的同时保持相同的延迟。这挑战了AI领域'能力与效率不可兼得'的传统观点,暗示了模型架构和推理算法的重大突破。

    10. The gains are especially strong in agentic coding, computer use, knowledge work, and early scientific research—areas where progress depends on reasoning across context and taking action over time.

      大多数人认为AI进步主要是在特定任务上的表现提升,但作者认为GPT-5.5的真正突破在于其跨上下文推理和长时间行动的能力,这挑战了人们对AI发展路径的传统认知。这种'代理式能力'的提升比简单的任务完成更为重要,因为它代表了AI向更接近人类工作方式的转变。

    11. We are treating the biological/chemical and cybersecurity capabilities of GPT‑5.5 as High under our Preparedness Framework. While GPT‑5.5 didn't reach Critical cybersecurity capability level, our evaluations and testing showed that its cybersecurity capabilities are a step up compared to GPT‑5.4.

      大多数人认为AI在网络安全领域的应用应该被严格限制或视为威胁,但作者认为GPT-5.5的网络安全能力是'进步'而非危险,并将其归类为'高级'而非'关键'风险级别。这与主流的'AI网络安全威胁论'相悖,暗示AI可能成为网络安全防御的重要工具而非主要威胁。

    12. GPT‑5.5 is priced higher than GPT‑5.4, it is both more intelligent and much more token efficient. In Codex, we have carefully tuned the experience so GPT‑5.5 delivers better results with fewer tokens than GPT‑5.4 for most users

      大多数人认为更强大的AI模型必然会导致更高的计算成本和资源消耗,但作者认为GPT-5.5虽然价格更高,但实际上更高效,能用更少的token提供更好的结果。这与AI领域'性能提升必然伴随成本上升'的共识相悖,暗示模型优化可能比规模扩张更经济高效。

    13. The viable path is trusted access, robust safeguards that scale with capability, and the operational capacity to detect and respond to serious misuse.

      大多数人认为随着AI能力增强,应该更严格限制其访问以防止滥用,但作者认为'可信任的访问'和'随能力扩展的安全保障'才是可行路径。这与主流的'限制性安全'观点相悖,暗示开放但有强监管的AI部署可能比封闭式AI更安全有效。

    14. GPT‑5.5 is our strongest agentic coding model to date. On **Terminal-Bench 2.0,** which tests complex command-line workflows requiring planning, iteration, and tool coordination, it achieves a state-of-the-art accuracy of 82.7%.

      大多数人认为AI在复杂编程任务中仍需要人类监督和干预,但作者认为GPT-5.5已经能在复杂的命令行工作流中达到82.7%的准确率,这挑战了'AI编程助手仍处于辅助阶段'的共识,暗示AI可能在某些编程领域已经接近或达到专业人类水平。

    15. GPT‑5.5 delivers this step up in intelligence without compromising on speed: larger, more capable models are often slower to serve, but GPT‑5.5 matches GPT‑5.4 per-token latency in real-world serving, while performing at a much higher level of intelligence.

      大多数人认为更强大的AI模型必然会牺牲速度和效率,但作者认为GPT-5.5打破了这一传统权衡关系,实现了更高智能的同时保持相同延迟。这挑战了AI领域'更大模型必然更慢'的共识,暗示模型架构优化可能比单纯扩大规模更重要。

    1. Jeremy didn't get laid off. He got leveraged.

      大多数人认为在裁员潮中,高额使用AI工具的员工可能会被视为成本负担而被裁掉,但作者提出了一个颠覆性的观点:像Jeremy这样大量使用AI工具的员工不仅没有被裁员,反而获得了更大的杠杆效应和影响力。这挑战了人们对AI成本与价值的传统认知。

    2. A US lab would never; well, unless you count a code red or Meta's throw money at the problem moves.

      大多数人认为美国AI实验室会始终保持技术领先优势并公开承认自己的不足,但作者暗示美国实验室(尤其是Meta)只会通过大量投入资金来掩盖技术差距,而非公开承认落后。这种观点挑战了人们对美国科技企业透明度和创新能力的传统认知。

    1. The tradeoff is that the same input can map to more tokens—roughly 1.0–1.35× depending on the content type. Second, Opus 4.7 thinks more at higher effort levels, particularly on later turns in agentic settings. This improves its reliability on hard problems, but it does mean it produces more output tokens.

      大多数人认为AI模型升级应该提高效率,减少资源消耗。但作者指出Claude Opus 4.7实际上会产生更多输出token,消耗更多计算资源。这种'效率降低'换取'可靠性提高'的权衡挑战了人们对AI发展必然带来效率提升的认知,表明在某些场景下,模型可能需要更多思考才能达到更好的结果。

    2. Our alignment assessment concluded that the model is 'largely well-aligned and trustworthy, though not fully ideal in its behavior'. Note that Mythos Preview remains the best-aligned model we've trained according to our evaluations.

      大多数人可能会认为最新、最强大的AI模型应该在对齐和安全性方面表现最好。但作者明确指出,虽然Claude Opus 4.7功能强大,但在对齐方面反而不如之前的Mythos Preview模型。这一反直觉的结论挑战了'能力越强,对齐越好'的普遍假设,暗示AI发展可能存在能力与对齐之间的权衡。

    3. On some measures, such as honesty and resistance to malicious 'prompt injection' attacks, Opus 4.7 is an improvement on Opus 4.6; in others (such as its tendency to give overly detailed harm-reduction advice on controlled substances), Opus 4.7 is modestly weaker.

      大多数人认为AI模型的每个新版本都应该在所有安全指标上都有进步。但作者明确指出Claude Opus 4.7在某些安全方面反而比前代模型表现更弱,这挑战了人们对AI安全线性进步的假设。这种非线性的安全表现表明,模型能力的提升可能伴随着某些方面的权衡,而非全面增强。

    4. Opus 4.7 is better at using file system-based memory. It remembers important notes across long, multi-session work, and uses them to move on to new tasks that, as a result, need less up-front context.

      大多数人认为AI模型在长对话中会逐渐'忘记'早期信息,需要不断重复上下文。但作者认为Claude Opus 4.7能够跨会话记忆重要信息,这挑战了人们对AI短期记忆局限的认知。这种持久记忆能力意味着AI可以真正进行长期项目,而不需要用户不断重复提供背景信息。

    5. Interestingly, this means that prompts written for earlier models can sometimes now produce unexpected results: where previous models interpreted instructions loosely or skipped parts entirely, Opus 4.7 takes the instructions literally.

      大多数人认为AI模型应该越来越能理解用户的意图,即使指令表达不够精确也能灵活处理。但作者认为Claude Opus 4.7反而更严格地遵循字面指令,这可能导致用户为旧模型编写的提示产生意外结果。这种'过度遵从'实际上是一种反直觉的进步,因为它减少了模型对用户意图的推测,增加了可预测性。

    1. Cursor counted the entire file as AI, even though we can see from the diff that it left plenty of the lines unchanged.

      大多数人认为AI代码指标应该精确追踪实际修改的代码行,但作者发现Cursor会将整个文件标记为AI生成,即使只修改了其中部分行,这表明AI工具的追踪系统存在严重缺陷,可能导致完全错误的贡献报告。

    2. customers should expect PCW values of 85%+, often 95%+. This is not a hallucination and is accurate given how we compute this metric

      大多数人认为AI代码生成工具应该客观、准确地衡量其贡献,但作者认为这些工具的报告数据被设计得极度偏向高AI贡献比例(85%-95%),因为它们的计算方法有严重缺陷,如不计算用户粘贴的代码、不计算自动添加的符号等,这些偏差导致AI贡献被高估。

    1. placing constraints upon them not only helps users and services build trust in them, but it also helps people more easily conceptualise what they do.

      大多数人认为限制AI代理的能力会限制其创新和价值,但作者认为约束实际上能建立信任并帮助用户理解功能。这个观点挑战了'无限制创新'的主流科技叙事,暗示适当的约束可能带来更大的价值和采用。

    2. lack of a well-defined user agent role in AI that's backed up by transparent, public standards... leaves a gap – it makes it harder for a marketplace to form.

      大多数人认为AI代理的主要问题是技术或安全方面,但作者认为缺乏明确定义的用户代理角色和透明标准才是根本问题,这阻碍了健康市场的形成。这个观点挑战了行业对AI发展的主流叙事,强调了制度架构比技术实现更重要。

    3. Every time you use an Internet-connected computer, you're trusting someone (and most likely, a multitude) to act on your behalf.

      大多数人认为互联网设备是工具,应该按照用户意图工作,但作者认为现代互联网设备实际上是代理,代表多方利益,这些利益可能与用户不一致。这挑战了我们对数字工具本质的理解,暗示我们使用的每台设备都在进行某种形式的'集体谈判'。

    1. A LeadDev survey found 54% of engineering leaders believe AI copilots will reduce junior hiring long-term.

      大多数人认为AI会创造新的就业机会,但作者引用调查表明,行业领导者实际上计划减少初级岗位招聘。这与AI创造就业的主流叙事相悖,揭示了AI可能导致的就业结构变化。

    2. When juniors skip debugging and skip the formative mistakes, they don't build the tacit expertise. And when my generation of engineers retires, that knowledge doesn't transfer to the AI.

      大多数人认为AI可以替代人类学习过程,但作者认为跳过调试和错误经验会阻碍隐性知识的形成,导致关键能力无法传承。这与AI可以完全替代人类学习的普遍认知相悖。

    3. A nuclear weapons program lost the ability to make a material it invented. The knowledge existed only in people, and the people were gone.

      大多数人认为技术文档和记录足以保存知识,但作者通过Fogbank案例表明,关键知识往往只存在于人的经验中,一旦相关人才流失,即使有文档也无法重建。这挑战了文档化足以保存知识的普遍认知。

    1. memory-driven experience scaling represents a crucial new frontier for agent scaling

      大多数人认为智能体扩展应该主要通过增加模型参数或计算资源来实现。但作者提出经验驱动的记忆扩展是智能体扩展的关键新前沿,这挑战了传统扩展范式,暗示未来的AI发展可能更关注如何有效利用经验而非仅仅是扩大规模。

    2. simple checklists evolved into memories with compositional, preventative logic structures

      大多数人认为智能体的学习过程是线性的,从简单到复杂逐步发展。但作者观察到智能体的记忆结构经历了质变,从简单的程序清单演变成具有组合性和预防性逻辑的复杂结构。这挑战了对AI学习过程的线性理解,表明学习可能是非线性的、突变的,而非渐进的。

    3. existing TTS methods often discard the exploration trajectory and treat the final answer as the only useful outcome

      在测试时扩展(Test-time scaling)领域,主流观点认为只有最终结果才是有价值的,探索过程只是达到结果的手段。但作者认为被忽视的探索轨迹实际上是一个丰富的数据源,可以加速智能体从经验中学习的能力。这一观点挑战了传统TTS方法的价值评估标准。

    4. this self-judgement does not need to be perfectly accurate, as we find ReasoningBank to be quite robust against judgment noise

      大多数人认为智能体的自我评估需要高度准确才能有效学习,因为错误的判断会导致错误的记忆形成。但作者认为即使自我判断存在噪声,ReasoningBank仍然能够有效运作,这挑战了传统对评估精确性的严格要求,表明系统可能比预期更能容忍不完美的自我评估。

    5. by over-emphasizing successful experiences, they miss out on a primary source of learning — their own failures

      主流观点认为成功经验是学习的主要来源,应该被优先记录和分析。但作者认为失败经验实际上可能是更重要的学习资源,因为它提供了反事实信号和潜在陷阱的宝贵信息。这一观点挑战了传统只关注成功案例的做法,提出失败可能是更强大的学习驱动力。

    6. by recording detailed actions instead of tactical foresight, they fail to distill higher-level, transferable reasoning patterns

      大多数人认为记录详细的行动轨迹是智能体学习的最佳方式,因为这样可以保留完整的决策过程。但作者认为这种方法实际上阻碍了学习,因为它只关注具体动作而非可转移的高层次推理模式。这挑战了传统记忆存储的常识,表明简单记录所有交互并不等同于有效学习。

    1. When a Fugu model is allowed to call itself recursively, reading its own prior output as context and deciding whether to revise its coordination strategy, a new form of test-time scaling emerges.

      大多数人认为AI模型的能力主要取决于训练阶段,推理阶段只是应用已学知识,但作者提出Fugu模型可以在推理时通过自我递归调用实现能力扩展,这挑战了传统AI推理阶段的局限性,暗示小型模型可能通过自我迭代达到超越其初始能力水平的表现。

    2. A core conviction at Sakana AI is that the most capable AI systems will not be monolithic models scaled in isolation, but collections of specialized agents working together.

      大多数人认为更强大的AI系统必然是更大规模、更复杂的单一模型,但作者明确表示最具能力的AI系统将不是孤立扩展的单一模型,而是多个专业化代理的集合。这直接挑战了当前AI领域追求更大单一模型的共识,提出了一个根本不同的研究方向。

    3. Sakana Fugu coordinates pools of frontier foundation models to achieve state-of-the-art performance across coding, mathematics, scientific reasoning, etc.

      大多数人认为最先进的AI系统应该是单一的大型基础模型,但作者认为通过协调多个前沿基础模型组成的系统可以达到更好的性能。这挑战了当前AI行业追求更大单一模型的趋势,提出了一个多模型协作的替代路径。

    1. Chinese authorities have reportedly been pushing data centers and public computing projects to use more domestic chips, including through reported bans on foreign-made chips, sourcing quotas, and requirements to pair Nvidia chips with Chinese alternatives.

      大多数人认为中国芯片政策主要是市场驱动,但作者揭示了中国政府通过强制配额、禁令等行政手段推动国产芯片使用。这一观点挑战了'中国AI发展主要依靠市场力量'的共识,突显了国家战略在技术发展中的主导作用。

    2. DeepSeek does not appear to have fully moved beyond Nvidia. The company's technical report reveals that it is using Chinese chips to run the model for inference, but...appears to have adapted only part of V4's training process for Chinese chips.

      大多数人认为中国AI公司已经完全摆脱了对Nvidia的依赖,但作者认为DeepSeek V4仍主要依赖Nvidia芯片进行训练,仅在推理阶段使用中国芯片。这一观点挑战了'中国AI已实现完全自主'的主流叙事,暗示技术脱钩比表面看起来更为复杂。

    3. DeepSeek V4 exceeds them all on coding, math, and STEM problems, making it one of the strongest open-source models ever released.

      大多数人认为开源AI模型在性能上无法匹敌闭源商业模型,但作者认为DeepSeek V4在多个关键领域超越了其他开源模型,甚至与顶级闭源模型相当。这挑战了'开源必然意味着性能妥协'的行业共识,暗示开源模型正在迅速缩小与商业模型的差距。

    1. The filing cabinet keeps getting bigger. But a bigger filing cabinet is still a filing cabinet.

      大多数人认为通过扩大上下文窗口和检索能力可以解决AI的'记忆'问题,但作者认为这本质上只是让文件柜变大,而没有改变其本质。这个观点挑战了当前AI领域对'扩展上下文'的主流研究方向,暗示我们需要从根本上重新思考AI如何存储和处理信息,而不仅仅是扩大容量。

    2. The current separation between training and deployment is not just an engineering convenience – it is a safety, auditability, and governance boundary.

      大多数人认为训练和部署的分离只是工程上的限制,但作者认为这种分离实际上是必要的边界,关乎安全、可审计性和治理。这个观点挑战了AI社区中普遍认为的'模型应该能够持续学习'的共识,暗示开放模型参数更新可能带来严重的安全和治理问题。

    3. The intelligence lives in the static parameters, and the apparent capabilities change radically depending on what you feed into the window.

      大多数人认为AI模型的智能来自于其参数和输入内容的结合,但作者认为智能实际上完全存在于静态参数中,输入内容只是触发不同表现的开关。这个观点挑战了主流认知,因为它暗示模型本身是固定的,而变化仅来自于外部输入,这与我们通常认为模型能够通过输入'学习'的观点相悖。

    1. Without our safeguards in place (which we do to measure a model's raw capabilities), only Mythos Preview and Opus 4.7 completed more than half the tasks.

      大多数人认为高级AI模型在没有安全措施的情况下会自主执行复杂任务,但作者暗示即使是最先进的模型在没有人类指导的情况下也难以完成大多数任务。这挑战了AI自主性和能力的普遍认知,暗示AI可能比人们想象的更依赖人类监督。

    2. We also welcome feedback and input from third parties and industry experts. We're currently working with The Future of Free Speech (an independent think tank at Vanderbilt University), the Foundation for American Innovation, and the Collective Intelligence Project

      大多数人认为科技公司会独立制定AI政策并保持控制,但作者强调Anthropic积极寻求外部机构和专家的合作。这挑战了科技公司通常的封闭决策模式,暗示AI治理需要多方参与而非企业单方面主导。

    3. if AI models can answer these questions well (that is, accurately and impartially), they can be a positive force for the democratic process.

      大多数人认为AI在政治领域会带来偏见和操纵风险,但作者认为AI可以成为民主进程的积极力量,前提是它能准确且无偏见地回答问题。这挑战了主流对AI政治应用的担忧,暗示AI可能比传统信息渠道更可靠。

    1. Out of 28 paid and 400 free routers: > 9 injected malicious code into tool calls > 17 touched researcher-owned AWS credentials > 1 drained $500k from an Ethereum wallet

      大多数人认为付费API路由器比免费路由器更安全,但作者的研究表明即使是付费路由器也存在严重安全风险,因为无论付费与否,这些中间服务都有能力访问和操纵所有数据。这挑战了人们对'付费等于安全'的普遍认知。

    2. Some attacks only fired after 50 prior calls. Others activated only in auto-approve mode.

      大多数人认为安全威胁会立即显现,但作者认为许多攻击是经过精心设计的,会延迟激活或在特定条件下才触发,因为攻击者采用渐进式策略来避免被检测。这挑战了人们对即时威胁检测的假设。

    1. a lightweight surrogate trained on them can absorb a significant portion of future traffic at near-zero marginal inference cost

      大多数人认为模型替换会带来明显的质量下降或需要持续监督。但作者提出轻量级代理模型可以'吸收大量未来流量'且'边际推理成本接近零',这种近乎零成本的替代方式颠覆了传统模型替换的质量-成本权衡观念。

    2. On a 150-class benchmark, the surrogate fully replaces the teacher

      大多数人认为复杂分类任务需要大型模型才能处理,小型代理模型只能处理简单任务。但作者展示了一个150类复杂任务中,小型代理模型完全能够替代教师模型,这挑战了'越大越好'的主流认知,证明了高效路由的潜力。

    1. The DESIGN.md format is at version `alpha`. The spec, token schema, and CLI are under active development. Expect changes to the format as it matures.

      大多数人期望成熟的设计系统规范应该是稳定和向后兼容的。但作者明确表示DESIGN.md仍处于alpha阶段并预期会有重大变化,这挑战了设计系统应该高度稳定的主流认知,表明创新性工具可以采用更灵活的演进路径。

    2. Components map a name to a group of sub-token properties: ... Variants (hover, active, pressed) are expressed as separate component entries with a related key name.

      大多数人认为组件变体应该通过嵌套结构或条件逻辑来组织,这是现代UI框架的标准做法。但作者选择将每个变体表示为独立的组件条目,这种扁平化结构挑战了组件变体的传统组织方式,可能使某些复杂场景的维护变得更加困难。

    3. Unknown section heading | Preserve; do not error

      大多数人认为严格的格式规范应该拒绝未知或不合规的部分,以确保一致性。但作者选择保留未知标题而不报错,这表明设计系统应该允许扩展和进化,而不是被严格规范所限制,这是一种反直觉的开放性设计原则。

    4. A DESIGN.md file combines machine-readable design tokens (YAML front matter) with human-readable design rationale (markdown prose). Tokens give agents exact values. Prose tells them _why_ those values exist and how to apply them.

      大多数人认为设计系统应该完全由机器可读的配置文件定义,以确保一致性和自动化。但作者认为DESIGN.md格式需要同时包含机器可读的YAML前缀和人类可读的Markdown正文,因为人类提供的上下文和设计推理对AI理解设计意图至关重要,这挑战了纯配置驱动的设计系统理念。

    5. Unknown component property | Accept with warning

      大多数人认为设计系统应该严格限制和验证所有属性,以确保一致性和可预测性。但作者认为应该接受未知组件属性,但仅发出警告。这种方法挑战了传统设计系统必须严格控制所有方面的观念,提供了一种更为灵活的方法,允许创新和扩展,同时仍保持基本的结构和约束。

    6. Components map a name to a group of sub-token properties: ... Valid component properties: backgroundColor, textColor, typography, rounded, padding, size, height, width.

      大多数人认为组件应该被定义为完整的、独立的实体,包含所有必要的样式和功能。但作者认为组件应该被定义为对已有设计 tokens 的引用和组合,而不是独立的样式定义。这种方法挑战了传统的组件设计理念,强调了设计系统中的复用性和一致性而非组件独立性。

    7. A DESIGN.md file combines machine-readable design tokens (YAML front matter) with human-readable design rationale (markdown prose). Tokens give agents exact values. Prose tells them _why_ those values exist and how to apply them.

      大多数人认为设计系统应该完全由机器可读的代码或配置文件定义,以确保一致性和自动化。但作者认为,将人类可读的设计 rationale 与机器可读的 tokens 结合是更好的方法,因为 prose 能提供设计意图和上下文,这对于 AI 理解和应用设计系统至关重要。这是一种将人类设计师的意图与机器执行能力相结合的非传统方法。

    1. At our request, the underwriters have reserved up to _______% of the shares of Class A common stock offered by this prospectus for sale at the initial public offering price through a directed share program to certain persons identified by our management and certain long-tenured employees, which may include parties with whom we have a business relationship and friends and family of management and such employees.

      大多数人认为IPO分配应该基于市场机制和机构投资者需求,但Cerebras预留大量股份给管理层、员工及其关系网络。这挑战了IPO公平分配的普遍认知,暗示公司可能优先考虑内部人利益而非最大化股东价值。

    2. We have applied to list our Class A common stock on the Nasdaq Global Select Market under the symbol 'CBRS,' and this offering is contingent upon the listing of our Class A common stock on the Nasdaq Global Select Market.

      大多数人认为IPO成功是公司财务健康的标志,但Cerebras将上市成功与股票上市直接挂钩,暗示公司可能认为即使融资成功,若不能在纳斯达克挂牌,其价值主张将大打折扣。这挑战了IPO过程中融资和上市是两个独立步骤的常规认知。

    1. Distributed systems background - real-time sessions, cloud infrastructure (AWS), and production reliability

      大多数人认为游戏引擎开发主要关注客户端性能和用户体验,但这里强调分布式系统、实时会话和云基础设施,表明ARC Prize将游戏视为分布式系统的一部分,这与传统游戏开发中客户端优先的理念形成鲜明对比。

    2. Hands-on experience building or maintaining a game engine (must), with strong Python fundamentals (must)

      大多数人认为高性能游戏引擎必须使用C++等低级语言,但这里明确要求Python作为游戏引擎的核心语言,挑战了游戏开发领域的传统认知,表明在AI评估场景中,开发速度和灵活性可能比性能优化更重要。

    3. A senior engineer to own and evolve the game engine and real-time play infrastructure behind the ARC-AGI series.

      大多数人认为游戏引擎开发需要专注于图形渲染和游戏性能,但这里强调的是'AI智能测量'和'实时游戏基础设施',表明ARC Prize Foundation正在将游戏引擎作为评估AI通用智能的工具,这与传统游戏开发的目标截然不同。

    1. Begin with fast checks that explain behavior, then add slower, heavier checks only when they reduce risk.

      大多数人可能认为全面的评估应该从一开始就包含所有可能的检查,但作者主张渐进式评估方法,从快速检查开始,只在必要时添加更复杂的检查。这个观点挑战了'一次性全面测试'的常规做法,主张风险驱动的评估策略。

    2. The most reliable way to improve a skill over time is to evaluate it the same way you would any other prompt for LLM applications.

      大多数人可能认为AI代理技能需要特殊的、独特的评估方法,但作者认为它们应该被视为普通LLM提示应用的一部分进行评估。这个观点挑战了AI代理评估需要特殊框架的共识,主张统一的方法论。