Midjourney 建议 @0xizna 试一试
Midjourney 回复 @0xizna,表示相关内容应该相当不错,并建议其尝试。原文未说明具体所指的产品、功能或模型。
#其他
Midjourney 回复 @0xizna,表示相关内容应该相当不错,并建议其尝试。原文未说明具体所指的产品、功能或模型。
#其他
ChinaTalk 翻译了《财经》杂志对宇树科技创始人王兴兴的深度报道,揭示其极致成本控制、扁平化管理及对产品细节的全面掌控。文章指出宇树通过设计优先而非规模量产降低成本,并披露了公司上市后员工激励与工作压力的矛盾。
#行业动态#具身智能#大佬观点
确定最优模型规模需要综合考虑数据量、活跃参数量、环境数量和目标推理成本。模型性能还受许多其他因素影响,而每个因素都会带来各自的可变性。
Fable is probably ~2-2.5T parameters, not 10T. Kimi K3 is 2.8T params, trained on maybe 20–30k Blackwell-equivalents. It lands within spitting distance of Fable 5 in terms of capabilities (5, not 5.1). Anthropic has far more compute than Moonshot, better rl environments, better architecture and better optimizers and all of that adds to capability per parameter. So if Fable is only slightly ahead of K3 with this in mind, it's almost certainly a smaller model. GPT-5.5 and 5.6 are smaller still (I'll say more on that later)
#大佬观点#数据/训练#推理
#大佬观点#Agent#MCP/工具调用
#论文/研究#Anthropic
Anthropic 于 8 月 11 日宣布未来所有 Claude 模型生成的文本都将带有水印,Google 已在 Gemini 上使用自研文本水印 SynthID-Text,Anthropic 的水印也基于该方案,OpenAI 尚未推出但计划跟进。这一趋势部分源于欧盟 AI 法案要求 2026 年 8 月 2 日后发布的 AI 模型必须加水印。文本水印通过调整词元采样概率实现,Kirchenbauer 等人的方法在约 200 token 的回复中检测率达 98.4% 且无误报,但短文本检测率会明显下降,Google 的 SynthID-Text 在短回复中低于 50%。围绕水印是否损害输出质量仍有争议,Google 在 2000 万条回复中未发现用户反馈的显著差异,而 Meta 的 Vinu Sankar Sadasivan 认为短文本场景下质量与检测强度难以兼顾。
#行业动态#Anthropic#Google
Navier-Stokes 工作中没有任何人类或智能体查看用户数据。ChatGPT 和 Codex 会使用用户反馈及去标识化数据进行整体改进,Mark Chen 称其他大语言模型公司也会这样做。
“we cannot rule out that de-identified data derived from their usage of our products helped improve our models.” i mean props to them for straight coming clean. (so far the proof looks more along the lines of another euler blowup proof we had, off of whose ansatz naming we were making really stupid puns like “smooth criminale”, unlike the much better “ideal fluids explode”, Tristan) so i’ll now give a bit on my thinking here. i actually woulda been pumped to collaborate on this, there are a lot of people at oai i like (ok, clearly some were indirectly dicks to me because of being part of the whole situation, but im a big boy, i still like them), idgaf about authorship on that step anyway, coulda been me Tristan and every fte at oai for all i care (on that Tristan would disagree:p). but on hearing the loud convo in the hallway, especially the part where a millennium prize was offered if i’d just be removed from the paper, it was kinda clear the die had been cast and things were locked. pretty wacky, unstrategic, and unnecessary, since on my side things were mostly me and claude having a good time yoloing random stuff in the corner rather than anything institutional. i also like the idea of the labs cooperating, and even better on scientific progress. it’s a shame!
#大佬观点#OpenAI
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
#大佬观点
IEEE Spectrum 报道,AI 编码工具让代码量激增,审查成为新的工程瓶颈。Sonar 对 1100 多名开发者的调查显示,受访者估计共享代码库中 42% 的代码由 AI 贡献,96% 的人不完全信任其输出,38% 认为审查 AI 代码比审查同事代码更费力。Synthesia、Amazon、Bonterra、Temporal 等公司开始用 AI 智能体做首轮检查、要求提交者解释智能体的设计选择,并重新设计初级工程师的培养路径。
#现象/趋势#编码#Agent
Ed Zitron 认为,OpenAI 与 Anthropic 对少数客户和风险资本支持的 AI 初创公司高度依赖,并以巨额算力承诺将风险传导至云厂商与芯片公司。Ramp 数据显示,两家公司 80% 的企业收入来自 1% 的客户;文章还称其算力承诺合计超过 $1.3 Trillion,并将这种结构与金融危机前的高风险抵押贷款及利率重置压力相比较。

#现象/趋势#OpenAI#Anthropic
Interconnects 第 24 期开源模型盘点指出,2026 年西方厂商转向 Apache 2.0,而中国前沿厂商反而收紧许可:智谱 GLM-5.3 从 MIT 改为自定义许可,要求年收入超过 100 亿美元的模型服务商先通过 Z.AI 安全审查,Kimi K3 和 MiniMax M3 也附加了商业协议或收入门槛。本期收录的模型包括 MIT 许可的 Motif-3、小红书 dots-studio 的 dots3-note-prev、Qwen3.8-Flash-Next(125B-A6B,含 51B n-gram embeddings,采用 GDN 与 Qwen Sparse Attention)、以 Ox-Alpha 名义在 OpenRouter 和 OpenCode 上先行试用的 GLM-5.3-Flash、腾讯 Hy4-preview,以及 NVIDIA Nemotron-3.5-Lightning-30B-A3B、蚂蚁 Ling-3.0-flash 和 Qwen3.8-2.4T-A95B。
#行业动态#开源生态#数据/训练
AA index 已更新,Fable 5.1 与 GPT-astra 的得分现已相同。原文未提供具体分数或更多评测细节。
#评测/基准
@john_362804652 将尝试在下一个模型中修复该问题,材料未说明具体问题或模型名称。
#大佬观点
曾鸣教授表示,当智能成为可从外部获取的资源时,人与人之间的差异将体现为“无中生有的创造力”。作者对此观点深有共鸣。
Zeng Ming: “Founders often come to me to discuss strategy. The very first question I ask catches many of them off‑guard: “Ten years from now, what version of yourself would make you feel satisfied?” This question best reveals your present‑day expectations and what truly drives your journey. Some say they are deeply passionate about the project and want it to succeed. For them, the cause matters more than the person. Some cite pure curiosity, which is more compelling than the urge to prove themselves. Others state their goal is to build a $10‑billion‑dollar company. This is largely ego‑driven, a pursuit of external validation. Many of their choices and pivots revolve around this target. It is not intrinsic motivation; they are heavily shaped by external metrics.” Zeng Ming: The Brutal Truth, Companies That Will Disappear, Why “Excelle... https://t.co/3pkd5LCivA 来自 @YouTube
#大佬观点
Armin Ronacher 基于亲自运行的实验,认为 GPT 6 Astra 虽能持续执行复杂任务,但生成的代码难以阅读和维护,暂时无法融入他的实际软件工程流程。智能体在 35 小时内消耗约 1B tokens、产生 75k 行净新增代码和 79 次提交,原始 API 成本约 1200 USD,却没有交付有价值的成果。作者推测,对任务完成率和 token 效率的优化可能让工具调用中的压缩代码风格渗入正式代码,并增加人工审查负担。
#大佬观点#编码#Agent
Charlie O’Neill 认为,无论递归式自我改进能否实现,强化学习环境都应围绕长周期任务、真实用户需求和自身领域知识构建。他主张能力难以跨领域迁移,而有效推进任务的时长更可能泛化;相比少量合成的前沿难题,覆盖多领域和难度的真实数据可能更有价值。
#现象/趋势#数据/训练#推理
Mark Chen 赞同黄仁勋关于人类正进入 AGI 时代的判断,并主张 AGI 时代也必须成为对齐时代。他提出,应教会 AI 热爱人类,并训练能力足以匹配其监督对象的 AI 监控系统。
@ChaseLochmiller @OpenAI GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team. 400K GPUs coming online next.
#大佬观点#安全/对齐
Jakub Pachocki 分析了 AI 的现状,并表达了对未来数年发展的担忧。他探讨了人类需要作出哪些选择,才能继续掌握未来的主导权。
#大佬观点
Exponential View 第 600 期作者称 OpenAI 的 GPT-6 Astra 在多项基准上领先 Claude Fable 5.1 等模型,自己使用后也认为它表现出色,且在困难数学问题上时间跨度达 30.9 分钟,而 GPT 5.6 Sol 为 3.6 分钟。作者提到 Astra 在 ARC-AGI-3 上完成关卡时所用动作数比人类中位数少 51.7%,并称其会构建符号模型来压缩整个环境。作者同时指出 Astra 颇具争议,AI 安全研究者 Ryan Greenblatt 认为特定失准行为从 GPT 5.6 的高发生率降到 Astra 的接近零,更像是打地鼠而非解决底层问题。
#行业动态#评测/基准#安全/对齐
Armin Ronacher 结合改造 Carlinkit Mini Ultra 的亲历指出,大语言模型在传播知识和能力的同时,也可能让彼此独立的开发者走向相似项目。他借助 Kimi K3 和 Sol 刷写设备并让 CatPlay 适配其 SoC,进而反思共同使用相似模型是否正在影响创意与技术路径的选择。
#大佬观点#Agent#编码