大模型好不容易学会数r，结果 换个字母就翻车了 ？ 而且还是最新的GPT-5。 杜克大学教授Kieran Healy表示，自己 让GPT-5数了数blueberry里有几个b，结果GPT-5斩钉截铁地回答3个 。 抓马的是，GPT-5刚发的时候还有网友让它数过blueberry里的r，结果数对了。 克雷西 发自 凹非寺 量子位 | 公众号 QbitAI 虽然博主想到了换掉strawberry，却没成想让GPT-5变得"没有B数"的，竟然不是单词而是字母…… 看来香槟还是开得早了亿点点啊（手动狗头）。 越不过的"蓝莓山" Healy撰写了一篇名为 "blueberry hill" （蓝莓山）的博客，展示了他和GPT-5之间针对"blueberry里有几个b"展开的一场"拉锯战"。 除了开头直接提问的结果外，Healy还尝试多次变换提示词策略，结果GPT-5的嘴都是比煮熟的鸭子还硬。 比如让它展示出这些b都是在哪里出现的，GPT-5明目张胆地把开头blue中的b数了两遍。 一看不奏效，Healy接着追问说，把这3个b给我拼出来，拼出来就可以。 结果拼是拼出来了，但是GPT-5还是坚持有三个b，并表示第三个b是第七 ...

GPT-5 is finally here. "Today we are finally releasing GPT-5," but the error-filled press conference was followed by ridicule. Many people hate GPT-5. It still doesn't have AGI (artificial general intelligence) . GPT-5 didn't bring (AGI) , which is disappointing. The whole release felt like it was pushed. It might be because they are in a hurry to commercialize it. It's a reflection on the AI parameter Scaling Law's "hard work makes miracles" hitting a wall. Scaling Law has indeed hit a wall. Hello everyone ...

2025.08.09 本文字数：2684，阅读时长大约4分钟 作者 |新皮层NewNewThing 王杰夫 8月8日凌晨，OpenAI终于发布了新一代有整数编号的GPT模型GPT-5，距上一代GPT-4发布已经过去2年 4个月零24天。 过去，每一代GPT模型都标志着某种技术上的突破：随着参数规模扩大，GPT-3「涌现」出了GPT-2没 有的智能水平；到了GPT-4，模型开始具有图像理解相关的多模态能力。相比之下，打磨了2年之久的 GPT-5却显得有些「平庸」：OpenAI说GPT-5是个「博士」，但在各项能力上，除了更低的幻觉——比 GPT-4o低约45%，比OpenAI o3低约 80%，GPT-5没有展示出先前模型没有的能力，AGI也没有到来。 幻觉降低是模型最大优化。 甚至连OpenAI自己都不再将GPT-5称作「模型」，而是将其定义为「一个统一的系统」（One unified system）。 报告中还提到，只有11%的企业表示在过去一年中更换过供应商。考虑到更换模型供应商现象是如此罕 见，那么OpenAI与Anthropic在短短半年内地位的颠倒只能说明，OpenAI在获得新的B端客户上大大落 后 ...

Core Insights - MiroMind aims to build a self-aware digital agent ecosystem, focusing on the continuous evolution of Artificial General Intelligence (AGI) through community collaboration and open-source principles [2][4]. Group 1: Open Source Ecosystem - MiroMind has developed a comprehensive open-source ecosystem that includes the Agent framework (MiroFlow), models (MiroThinker), data (MiroVerse), and training infrastructure (MiroTrain/MiroRL), all of which are open for learning, reuse, and further development [1][8]. - The MiroFlow framework achieved a state-of-the-art (SOTA) score of 82.4 on the GAIA validation set, surpassing existing commercial model APIs [1][12]. - MiroThinker, the core model, reached a SOTA performance of 60.2% on the GAIA-Text-103 dataset, nearing the performance level of OpenAI's Deep Research [1][15]. Group 2: Community Collaboration - MiroMind fosters a developer-centric environment that encourages community participation through data requests, feature customization, and technical challenges, with feedback directly influencing project development [2][22]. - The project organizes various community activities such as competitions, leaderboards, and hackathons to enhance developer engagement and contribution [22]. Group 3: Key Personnel - The project is led by Chen Tianqiao, a renowned entrepreneur known for his strategic vision and significant contributions to brain science and AI [4]. - Dai Jifeng, a key figure in the project, is a professor at Tsinghua University with extensive experience in computer vision and deep learning, having published over 80 papers with significant citations [5][6].

以下文章来源于AI产品榜 ，作者李榜主 AI产品榜 . AI产品榜 aicpb.com 按月发布AI产品榜单。AI产品榜大会，是你必参的会。 发起人：李榜主 wx：QBB2378 第26期AI产品榜·网站榜(Web) （2025年07月） 第26期AI产品榜·网站榜(Web)（2025年07月）由AI产品榜、36kr、硅星人｜沃垠AI联名发布。 AI产品榜2025年07月网站榜，本文里包含19个AI榜单。 | AI产品榜·网站榜(web) | | | --- | --- | | 全球总榜 | 全球搜索引擎 | | 国内总榜 | 全球聊天机器人 | | 出海总榜 | 全球Al虚拟角色 | | 全球 · 增速/降速榜 | 全球AiPPT工具 | | 国内·增速榜 | 图片生成/编辑 | | 全球视频生成/编辑 | 全球音乐/会议助理 | | Al代码助理 HOT | Al云榜 | | AI产品榜·智能体榜 HOT | | 非商用引用数据标注来源：【公众号@AI产品榜aicpb.com】 ChatGPT上线不到3年 跻身全球第5大网站 全球五大网站分别是：上线距今27年搜索引擎Google、20年的视频分享分平台Y ...

If AGI arrives, Nvidia won't sell chips. It will sell cognition itself.At 40 times forward earnings, Nvidia (NVDA 0.60%) looks expensive through a traditional semiconductor lens. But that framework collapses if artificial general intelligence (AGI) arrives by 2030, as OpenAI, Anthropic, and other labs privately expect. Nvidia won't just supply artificial intelligence (AI) infrastructure. It could collect a toll on every intelligent operation on the planet.Think of it as the intelligence toll: a per-cycle fe ...

北京时间8月8日凌晨，OpenAI发布了它们最新一代的GPT模型——GPT-5。 | | GPT-5 | Gemini 2.5 | Grok | Claude 4.1 | | --- | --- | --- | --- | --- | | | (high) | Pro | 4 | Opus | | AIME '25 (no tools) | 94.6% | 93.8% | 90.5% | 94.1% | | FrontierMath (with python tool | 26.3% | 27.1% | 24.0% | 25.8% | | only) | | | | | | GPQA diamond (no tools) | 85.7% | 86.1% | 83.2% | 85.9% | | HLE[1] (no tools) | 24.8% | 23.5% | 21.1% | 24.2% | | HMMT 2025 (no tools) | 93.3% | 92.9% | 89.7% | 93.0% | GPT-5以个位数优势领先竞争对手 这种合成数据的新应用，让前一代先进模型生成高质量数据，让后 ...

Core Viewpoint - The release of GPT-5 marks a significant evolution in AI capabilities, transitioning from a focus on conversation to practical applications, with a unified intelligent system designed to handle various tasks efficiently [6][19]. Group 1: GPT-5 Features and Architecture - GPT-5 introduces a unified intelligent system that includes a fast model for general queries, a deep reasoning model for complex problems, and a real-time router to dynamically select the appropriate model based on user input [7][9]. - The model supports an input limit of 272,000 tokens and an output limit of 128,000 tokens, accommodating both text and image inputs [9]. - OpenAI aims to phase out older models, signaling a shift towards a more cohesive and collaborative AI system [9][10]. Group 2: Performance Metrics - GPT-5 achieved impressive scores in various benchmarks, including 94.6% in the AIME 2025 math test and 74.9% in the SWE-Bench for software engineering tasks [16]. - Despite its strong performance, there were issues during the presentation, such as inconsistencies in benchmark data displayed [12][15]. Group 3: Market Strategy and Pricing - OpenAI's pricing strategy for GPT-5 is aggressive, charging only $1.25 per million input tokens, which is significantly lower than its predecessor GPT-4o and competitive against other models [21]. - This pricing strategy is intended to capture market share and foster a robust developer ecosystem [21]. Group 4: User Experience and Feedback - While general user engagement with GPT-5 has increased, professional users have expressed dissatisfaction with its writing capabilities compared to previous models [35][24]. - The model's reliability and ability to reduce hallucinations have been emphasized, with claims of improved performance in common use cases such as programming and writing [30][28]. Group 5: Future Implications - The release of GPT-5 signifies a shift towards a more mature and specialized phase in AI development, moving away from the initial excitement of rapid advancements [37]. - The industry may be entering a new era where the focus is on practical applications and reliability, particularly for developers and creative writers [38].

The All-In-One MMS will come pre-configured with OpenAI's newly released, high-performance large language models (LLMs), GPT-OSS-120B and GPT-OSS-20B.SINGAPORE, Aug. 7, 2025 /PRNewswire/ -- Super X AI Technology Limited (NASDAQ: SUPX) ("the Company" or "SuperX") today announced the official launch of its latest All-in-One Multi-Model Servers ("MMS"). As the first enterprise-grade AI infrastructure to support the dynamic collaboration of multiple models by SuperX, this MMS is centered on being out-of-the-box ...

GPT-5，曾经差点难产？这条诞生路，简直是烈火炼真金。一边是人才出走、小扎截胡、团队内部陷入混乱，另一边，推理模型魔咒让研究者苦恼不已， 项目甚至一度停摆。外媒曝出这期GPT-5诞生内幕，可谓亮点满满，干货十足。 就在刚刚，外媒The Information曝出了关于GPT-5的一大波最新内幕，众多猛料来了！ 比如，GPT-5并未取得技术突破，不存在GPT-3到GPT-4这种级别的跃升。 参与此轮融资的，有一大波全新投资者，其中Dragoneer投资集团以28亿美元领投本轮，Blackstone、TPG、Fidelity、Founders Fund、红杉资本等跟投。 不过，虽说Dragoneer是本轮融资的最大出资方，但软银仍是整个400亿融资计划的牵头者。 GPT-5还没发布，各方势力都下场了，这不免让人把期待值拉满，屏息等待下周的盛况了。 Orion陨落真相，GPT-5没做出来，降级成4.5了 比如，OpenAI正面临着严重的数据瓶颈和技术难题。 还有一个劲爆大瓜，OpenAI大波核心研究者一下子被小扎撬走，直接导致了OpenAI内部的组织架构混乱！ 为此，研究副总裁Jerry Tworek在Slack ...