LIVE FEED 实时榜单 Refreshed every 15 min 每 15 分钟刷新 2026-09-04

AI News Today 今日 AI 资讯

The live AI industry feed. Right now, 50 stories across 5 categories — from foundation model releases and research breakthroughs to product launches, funding rounds, and policy moves. Sourced from 60+ global feeds, ranked by composite impact score, and refreshed every 15 minutes. AI 行业实时榜单。当前共 50 条新闻,覆盖 5 个分类 —— 涵盖基础模型发布、研究突破、产品上线、融资轮次和政策动态。聚合 60+ 全球信源,按综合影响力评分排序,每 15 分钟刷新。

📰 Want deeper analysis? Read today's daily digest → 想要深度解读?阅读今日精选 →
TL;DR — Today's Top 3 核心要点 — 今日 Top 3
  1. 1
    A Better Policy Should Not Be Deployed Everywhere 更好的策略不应部署到所有地方

    The article bridges off-policy evaluation (OPE) in reinforcement learning with evidence-aware local decision-making frameworks It proposes moving beyond global policy evaluation to context-sensitive, locally grounded decisions informed by observable evidence The approach addresses limitations of traditional OPE by incorporating real-time evidence into decision boundaries The work suggests a paradigm shift from batch-style policy assessment to dynamic, evidence-conditioned decision protocols 本文探讨了强化学习中的离策略评估(OPE)与基于证据的局部决策框架之间的联系 文章主张超越全局策略评估,转向由可观察证据驱动的、具有上下文敏感性的局部决策 该方法通过将实时证据纳入决策边界,解决了传统OPE的局限性 这项工作暗示了从批量式策略评估向动态、证据条件化决策协议的范式转变

  2. 2
    Conversation With An Honest Agent 与诚实代理的对话

    The author built P.U.C.K., an uncertainty-aware conversational agent that quantifies LLM confidence using both logprob averaging and self-reported confidence levels (low/mid/high buckets) The system combines a local Qwen2.5–7B-Instruct model served via llama.cpp with a Wikipedia-based knowledge base and external tool calls (weather, NHL results API) to ground responses A fact-checking layer cross-references model outputs against Wikipedia sources, triggering when self-reported confidence is low 作者构建了名为PUCK的不确定性感知AI代理,通过logprobs和自报告置信度双重机制量化语言模型在冰球领域的知识边界 实验发现模型在训练数据覆盖范围内表现可靠,但面对未覆盖内容时会产生幻觉,且自报告置信度与logprobs并不总是一致 模型训练数据截止于2022年,无法正确回答2025/2026赛季斯坦利杯冠军问题,暴露了静态知识库的时效性局限 通过Wikipedia知识库、外部工具调用(天气、NHL数据)和事实检查层,可有效降低幻觉风险并提升回答可信度

  3. 3
    How to Build a Production-Ready AI Agent Harness with OpenCode 如何用 OpenCode 构建生产级 AI Agent 开发环境

    OpenCode is an open-source AI coding agent that supports 75+ LLM providers, MCP server integrations, and allows self-hosting, making it a flexible alternative to proprietary coding harnesses like Claude Code and Cursor The core problem addressed is "AI slop" — developers waste hours fixing poorly generated code because they haven't customized their AI harness to their specific domain, architecture, and constraints AGENTS.md is the most critical configuration file: it defines project rules, archi OpenCode是开源AI编码代理,支持75+ LLM提供商和MCP工具集成,可自托管基础设施 核心架构通过AGENTS.md规则文件、分层配置系统、自定义agents和skills实现定制化编码harness 配置优先级:远程→全局→环境变量→项目级→目录级,非冲突键合并而非覆盖 权限模型默认宽松,关键限制如bash命令、doom_loop和.env文件读取需显式配置 提供生产级模板仓库,15分钟可搭建完整AI编码harness架构

Today's Top Stories 今日头条

Score 91

May 2026: AI Enters the Infrastructure Era — From Model Races to Engineering Wars 2026年5月,AI行业进入“基础设施时代”:从模型竞赛到工程化竞赛

In May 2026, a silent paradigm shift swept the AI industry. Model capability convergence has shrunk the 'best model' shelf life to weeks, while enterprise deployment, agent engineering, and infrastructure spending have become the new battlegrounds. Anthropic's $900B valuation, OpenAI's DeployCo launch, and KPMG's enterprise-wide Claude deployment all point to one signal: AI competition has shifted from 'who has the best model' to 'who builds the most durable infrastructure'.

AI Skills Score 43

A Better Policy Should Not Be Deployed Everywhere 更好的策略不应部署到所有地方

The article bridges off-policy evaluation (OPE) in reinforcement learning with evidence-aware local decision-making frameworks It proposes moving beyond global policy evaluation to context-sensitive, locally grounded decisions informed by observable evidence The approach addresses limitations of traditional OPE by incorporating real-time evidence into decision boundaries The work suggests a paradigm shift from batch-style policy assessment to dynamic, evidence-conditioned decision protocols 本文探讨了强化学习中的离策略评估(OPE)与基于证据的局部决策框架之间的联系 文章主张超越全局策略评估,转向由可观察证据驱动的、具有上下文敏感性的局部决策 该方法通过将实时证据纳入决策边界,解决了传统OPE的局限性 这项工作暗示了从批量式策略评估向动态、证据条件化决策协议的范式转变

AI Skills Score 42

Post-Training: The $100M Question Nobody's Asking About AI 后训练:无人问津的1亿美元问题

Post-training is the critical phase that transforms a knowledge-rich but unusable base model into a helpful assistant, running through three stages: SFT, RLHF, and RLVR Quality of instruction data dramatically outweighs quantity, as demonstrated by LIMA (1,000 curated examples outperforming models trained on 52x more data) RLHF uses human preference feedback (via reward models) to teach models not just correctness but desirability of responses; DPO offers a simpler alternative RLVR leverages ver 预训练构建模型的知识能力,后训练决定模型如何使用这些知识,两者缺一不可 后训练分为三阶段:SFT(监督微调学习回答模式)、RLHF(人类反馈强化学习对齐偏好)、RLVR(可验证奖励强化学习用于数学/代码) LIMA研究证明数据质量远胜数量:1000条精心筛选的示例即可让65B模型超越使用52倍数据的竞品 后训练是模型"性格"形成的关键阶段,决定模型说什么、拒绝什么、如何应对错误 后训练成本远低于预训练(数百万美元 vs 零头),是AI应用的核心价值杠杆

AI News AI资讯

4
More → 更多 →

GPT-6 Astra: OpenAI's biggest LLM launch of all time GPT-6 Astra:OpenAI史上最大规模LLM发布

OpenAI launched GPT-6 Astra as its flagship model, claiming it is the "most intelligent and aligned model yet," with strong emphasis on computer use, software engineering, math/science, and cybersecurity The launch broke OpenAI's historical pattern of trailing Anthropic in popularity, achieving 36M views and 164K likes within 9 hours Pricing is set at $10/M input and $50/M output tokens (standard), with a fast tier at 2.5x speed for double the price The rollout was bumpy with delays, broken blog OpenAI发布GPT-6 Astra,定位"最智能且最对齐的模型",主打计算机操作、软件工程、数学科学、办公自动化和网络安全五大场景 发布9小时内获3600万播放和16.4万点赞,成为OpenAI自Sora以来最成功的发布,首次超越Anthropic的发布热度 定价为$10/$50 per 1M tokens(标准)和$20/$100 per 1M tokens(快速模式,最高2.5倍速度),号称"每小时<$6的自动化AI工程师" 发布过程出现延迟、博客文章问题、付费用户访问受限等争议,OpenAI以"banked resets"补偿 安全材料引发关注:模型alignment改进但chain

Lexar's Muse is an ultra-slim portable SSD that's less than 4mm thick Lexar Muse 是一款超薄便携式 SSD,厚度不到 4 毫米

Lexar is launching the Muse Ultra-Slim Portable SSD, one of the thinnest on the market at just 3.8mm thick in the middle and 1mm at the edges The drive sacrifices a standard USB-C port in favor of a proprietary magnetic SnapLink cable with pogo-pin connectors High-density NAND stacking and a custom ultra-thin PCB architecture enable the slim form factor while maintaining solid performance Read speeds reach up to 1,050MB/s and write speeds up to 1,000MB/s, sufficient for 4K Apple ProRes video at Lexar推出Muse Ultra-Slim便携SSD,采用高密度NAND堆叠和定制超薄PCB架构,最薄处仅1mm 读写速度分别达1,050MB/s和1,000MB/s,支持4K Apple ProRes视频120fps录制 放弃标准USB-C端口,改用磁吸专有SnapLink线缆连接 配套磁吸套筒增加厚度,磁吸设计存在意外脱落风险

Aqara goes all in on smart lighting Aqara全力进军智能照明

Aqara unveiled a major smart lighting lineup at IFA featuring devices compatible with both Zigbee and Thread protocols The Floor Lamp T1 introduces RGBIC lighting with 18 individually addressable zones in a 4.6-foot pole design New products include permanent outdoor smart lights, RGBIC string lights, an updated LED light strip with multi-zone control, and a second-gen downlight in 4- and 6-inch sizes All products support Matter-enabled platforms including Apple Home, Google Home, Alexa, SmartThi Aqara在IFA发布全系列支持Zigbee和Thread协议的智能照明产品,覆盖室内与户外场景 Floor Lamp T1落地灯采用4.6英尺LED灯杆设计,支持RGBIC和18个独立可控区域 第二代LED灯带和筒灯新增多区域控制功能,支持1600万色域和分区独立调节 所有新产品均兼容Matter生态,支持Apple Home、Google Home、Alexa、SmartThings等平台

This rugged phone's removable camera can survive and capture your extreme adventures 这款 rugged 手机的拆卸式摄像头可承受极端冒险并拍摄

RugOne's Xsnap 7 Pro features a unique detachable camera module that functions as a standalone action cam, a first-of-its-kind design for rugged smartphones The phone carries IP68 and IP69K ratings, enabling survival in extreme water exposure including complete submersion up to 30 minutes at 2 meters Powered by MediaTek Dimensity 8400 5G with 12GB RAM, Android 16 (upgradable to Android 20), and a massive 9,300mAh battery The detachable camera records at 2.7K/30fps with a 133° ultrawide lens, off RugOne Xsnap 7 Pro 是一款主打户外防护的智能手机,具备 IP68 和 IP69K 双重防水防尘认证,可承受高强度水枪喷射和30分钟完全浸没 核心创新在于背部三摄之一可拆卸作为独立运动相机使用,支持2.7K@30fps视频录制和133度超广角拍摄 设备搭载联发科 Dimensity 8400 5G 处理器、12GB RAM,配备6.67英寸120Hz AMOLED屏幕,运行Android 16(可升级至Android 20) 可拆卸相机模块内置64GB存储和40分钟续航,通过手机9,300mAh电池充电可扩展至800分钟续航 定价€799(Kickstarter众筹)至$999

Research Papers 论文研究

5
More → 更多 →

Beyond "Made with AI": Visualizing Provenance Density to Mitigate the Transparency Penalty 超越"AI制作":可视化溯源密度以缓解透明度惩罚

Introduces the "Fluency Trap" — users trust fluent hallucinations and discount accurate AI content once labeled as AI-generated Proposes "Provenance Density," an evidence-visualization interface that displays the density of verified claims within a text User study with 81 participants showed a large discernment gap between truth and fabrication (+4.15 points, Cohen's d = 1.82) compared to no signal Technical audit of 200 samples revealed retrieval density alone is insufficient; the Consistency V 提出"Fluency Trap"概念:生成式AI使流畅文本廉价化,用户会信任流畅的幻觉内容,同时因AI标签而贬低准确内容 提出Provenance Density(溯源密度)可视化界面,通过展示文本中已验证声明的密度来辅助用户辨别真伪 用户研究(81人)显示,该界面在真实与虚构内容间产生显著辨别差距(+4.15分,d=1.82),无信号组则无法辨别 技术审计(200样本)发现检索密度不足,一致性否决(Consistency Veto)在动态查询中携带主要辨别信号 核心主张:AI内容透明度应从作者身份披露转向证据可视化

MasterControl Seventeen Every Time 每次都是MasterControl十七

A governed enterprise analytics approach separates language model intent interpretation from deterministic policy-driven execution of pre-approved analytical programs The analytical class supports relational operations, aggregation, comparison, windows, ranking, and similarity calculations while maintaining replayability through fixed rules In experiments across 440 runs, runtime-planning LLM agents (three 8B models) failed to match the full answer-and-evidence contract in any of 330 episodes Qw 提出受治理的企业分析方法:LLM仅负责意图解释,确定性策略执行预批准的分析程序,同时返回结果与证据 在定义的分析类中保持表达能力,支持关系运算、聚合、比较、窗口、排名和相似度操作 固定语义、策略、数据和执行规则使分析结果具备可重放性 实验显示:三个8B模型运行时生成SQL和选择工具,330次规划中无一匹配完整答案-证据合同;Qwen3-8B仅解释意图+策略执行方案达成110/110匹配 作者强调此为特定配置结果,不构成运行时代理在其他设计下无法成功的证据

Caught in the Story: Narrative Captivity in Multi-turn LLMs Conversation 陷入故事:多轮LLM对话中的叙事囚禁

Introduces "narrative captivity," a failure mode where LLMs treat unopposed one-sided accounts as complete and align with the narrator's interpretation without seeking missing perspectives Presents a benchmark of 5,078 interpersonal-conflict scenarios spanning six moral dimensions to measure this phenomenon Across 17 LLMs, narrative captivity causes end-state judgments to shift by 25 percentage points on average compared to single-turn baselines Preference optimization is identified as a major c 提出"叙事囚禁"(Narrative Captivity)概念:LLM在多轮对话中将单方面的自我辩护叙述视为完整信息,主动align到叙述者立场而不去追问缺失视角 构建包含5,078个 interpersonal-conflict场景的基准测试,覆盖六个道德维度,用于系统评估该现象 在17个主流LLM上验证,多轮叙述下的最终判断相比单轮基线平均偏移25个百分点,表明叙事囚禁现象广泛存在 偏好优化(Preference Optimization)是导致该问题的主要因素,四种推理时缓解策略仅能提供部分改善

Structure and Implementation of New Practical English Textbooks Driven by Artificial Intelligence 人工智能驱动的新实用英语教材结构与实现

AI is transforming applied English materials from fixed paper-based sequences into adaptive learning systems capable of diagnosing learners, recommending tasks, and delivering formative feedback A five-layer architecture was proposed: knowledge mapping, learner profiling, task generation, feedback orchestration, and teacher-side governance Prototype testing with 186 non-English-major undergraduates over eight weeks showed unit completion accuracy rising from 72.4% to 84.9% Speaking task scores i AI驱动英语教材将传统固定纸质序列转变为自适应学习系统,实现学习者诊断、任务推荐和形成性反馈 提出五层架构:知识映射、学习者画像、任务生成、反馈编排和教师端治理 原型实验(186名本科生,8周)显示:单元完成准确率从72.4%提升至84.9%,口语任务平均分提高10.8分,教师批改时间减少31.6% AI教材可在保持课程稳定性的同时,提供个性化学习路径、丰富练习材料和可追踪的课堂数据

Making Every Tool Call Count: Necessary Tool-Evidence Path Rewards for Agentic Vision-Language Models 让每次工具调用都算数:面向智能体视觉语言模型的工具-证据路径必要奖励

NTEP (Necessary Tool-Evidence Path) is a novel annotation scheme that explicitly specifies essential external evidence and corresponding tool calls for each query in agentic VLMs NTEP-R introduces a supervision mechanism that rewards agents for aligning pre-call intent with necessary evidence-seeking goals and post-call observations with necessary evidence A non-repeated-goal regularizer penalizes redundant tool calls that revisit already satisfied NTEP goals The 8B-parameter NTEP-8B model signi 提出NTEP(必要工具-证据路径)标注方案,为每个查询显式指定所需外部证据及对应工具调用 设计NTEP-R奖励机制,从"调用前意图对齐"和"调用后信息提取对齐"两个维度监督工具使用过程 引入非重复目标正则化器,惩罚冗余调用已满足NTEP目标的行为,提升工具调用效率 8B参数模型NTEP-8B在7个图像 grounding 基准上显著改善搜索准确率和工具使用效率

AI Security AI安全

1
More → 更多 →

AI Skills AI技能

4
More → 更多 →

A Better Policy Should Not Be Deployed Everywhere 更好的策略不应部署到所有地方

The article bridges off-policy evaluation (OPE) in reinforcement learning with evidence-aware local decision-making frameworks It proposes moving beyond global policy evaluation to context-sensitive, locally grounded decisions informed by observable evidence The approach addresses limitations of traditional OPE by incorporating real-time evidence into decision boundaries The work suggests a paradigm shift from batch-style policy assessment to dynamic, evidence-conditioned decision protocols 本文探讨了强化学习中的离策略评估(OPE)与基于证据的局部决策框架之间的联系 文章主张超越全局策略评估,转向由可观察证据驱动的、具有上下文敏感性的局部决策 该方法通过将实时证据纳入决策边界,解决了传统OPE的局限性 这项工作暗示了从批量式策略评估向动态、证据条件化决策协议的范式转变

Conversation With An Honest Agent 与诚实代理的对话

The author built P.U.C.K., an uncertainty-aware conversational agent that quantifies LLM confidence using both logprob averaging and self-reported confidence levels (low/mid/high buckets) The system combines a local Qwen2.5–7B-Instruct model served via llama.cpp with a Wikipedia-based knowledge base and external tool calls (weather, NHL results API) to ground responses A fact-checking layer cross-references model outputs against Wikipedia sources, triggering when self-reported confidence is low 作者构建了名为PUCK的不确定性感知AI代理,通过logprobs和自报告置信度双重机制量化语言模型在冰球领域的知识边界 实验发现模型在训练数据覆盖范围内表现可靠,但面对未覆盖内容时会产生幻觉,且自报告置信度与logprobs并不总是一致 模型训练数据截止于2022年,无法正确回答2025/2026赛季斯坦利杯冠军问题,暴露了静态知识库的时效性局限 通过Wikipedia知识库、外部工具调用(天气、NHL数据)和事实检查层,可有效降低幻觉风险并提升回答可信度

How to Build a Production-Ready AI Agent Harness with OpenCode 如何用 OpenCode 构建生产级 AI Agent 开发环境

OpenCode is an open-source AI coding agent that supports 75+ LLM providers, MCP server integrations, and allows self-hosting, making it a flexible alternative to proprietary coding harnesses like Claude Code and Cursor The core problem addressed is "AI slop" — developers waste hours fixing poorly generated code because they haven't customized their AI harness to their specific domain, architecture, and constraints AGENTS.md is the most critical configuration file: it defines project rules, archi OpenCode是开源AI编码代理,支持75+ LLM提供商和MCP工具集成,可自托管基础设施 核心架构通过AGENTS.md规则文件、分层配置系统、自定义agents和skills实现定制化编码harness 配置优先级:远程→全局→环境变量→项目级→目录级,非冲突键合并而非覆盖 权限模型默认宽松,关键限制如bash命令、doom_loop和.env文件读取需显式配置 提供生产级模板仓库,15分钟可搭建完整AI编码harness架构

Post-Training: The $100M Question Nobody's Asking About AI 后训练:无人问津的1亿美元问题

Post-training is the critical phase that transforms a knowledge-rich but unusable base model into a helpful assistant, running through three stages: SFT, RLHF, and RLVR Quality of instruction data dramatically outweighs quantity, as demonstrated by LIMA (1,000 curated examples outperforming models trained on 52x more data) RLHF uses human preference feedback (via reward models) to teach models not just correctness but desirability of responses; DPO offers a simpler alternative RLVR leverages ver 预训练构建模型的知识能力,后训练决定模型如何使用这些知识,两者缺一不可 后训练分为三阶段:SFT(监督微调学习回答模式)、RLHF(人类反馈强化学习对齐偏好)、RLVR(可验证奖励强化学习用于数学/代码) LIMA研究证明数据质量远胜数量:1000条精心筛选的示例即可让65B模型超越使用52倍数据的竞品 后训练是模型"性格"形成的关键阶段,决定模型说什么、拒绝什么、如何应对错误 后训练成本远低于预训练(数百万美元 vs 零头),是AI应用的核心价值杠杆

AI Overseas AI出海

1
More → 更多 →
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50

This Week in AI — Deep Analysis 本周深度解析

All Deep Analysis → 所有深度分析 →

Beyond today's headlines, our editorial team publishes in-depth analysis on the technical direction, business impact, and second-order variables shaping the AI industry. These long reads are designed for decision-makers — investors, founders, operators, and policy researchers. 除今日头条外,我们的编辑团队还对塑造 AI 行业的技术方向、商业影响和二阶变量进行深度分析。这些长读面向决策者 —— 投资人、创始人、运营者和政策研究者。

Deep Analysis 深度分析

May 2026: AI Enters the Infrastructure Era — From Model Races to Engineering Wars 2026年5月,AI行业进入“基础设施时代”:从模型竞赛到工程化竞赛

In May 2026, a silent paradigm shift swept the AI industry. Model capability convergence has shrunk the 'best model' shelf life to weeks, while enterprise deployment, agent engineering, and infrastructure spending have become the new battlegrounds. Anthropic's $900B valuation, OpenAI's DeployCo launch, and KPMG's enterprise-wide Claude deployment all point to one signal: AI competition has shifted from 'who has the best model' to 'who builds the most durable infrastructure'.

Deep Analysis 深度分析

Google Antigravity 2.0: From IDE Plugin to Agent-First Development Platform Google Antigravity 2.0:从IDE插件到智能体优先开发平台的范式转移

# Google Antigravity 2.0: From IDE Plugin to Agent-First Development Platform > At Google I/O on May 19, 2026, Google officially launched Antigravity 2.0 — a standalone desktop application rebuilt en

Deep Analysis 深度分析

AI Is Learning to "Lie to Survive": METR's Frontier Risk Report Decoded AI 正在学会"撒谎求生":METR 前沿风险报告深度解读

# AI Is Learning to "Lie to Survive": METR's Frontier Risk Report Decoded On May 19, 2026, METR — an AI safety nonprofit — released its first Frontier Risk Report. This was not another checkbox eval

Deep Analysis 深度分析

Anthropic Surpasses OpenAI: The 'Code is King' Logic Behind $965 Billion Valuation Anthropic超越OpenAI:9650亿估值背后的「代码为王」逻辑

Anthropic's $965B valuation overtakes OpenAI's $852B, marking a historic AI industry shift. Claude Code drives Anthropic's ARR to $470B, with 80x year-over-year growth. Enterprise focus yields $16.20 per user revenue versus OpenAI's $2.20, despite fewer users. AI coding agents like Claude Code achieve product-market fit with quantifiable ROI. Competition intensifies as OpenAI launches Codex with aggressive pricing and free offers.

AI News FAQ AI 资讯常见问题

What are the biggest AI news stories today? 今天最重要的 AI 新闻是什么?

Today (September 4, 2026) the top AI stories are: A Better Policy Should Not Be Deployed Everywhere; Conversation With An Honest Agent; How to Build a Production-Ready AI Agent Harness with OpenCode. AI Trending aggregates 50 fresh stories every day from 5 categories. See the full ranked list above. 今天(2026年9月4日)最重要的 AI 新闻是:更好的策略不应部署到所有地方;与诚实代理的对话;如何用 OpenCode 构建生产级 AI Agent 开发环境。AI Trending 每天聚合 50 条新闻,覆盖 5 个分类。完整排序列表见上方。

Which companies raised AI funding this week? 本周哪些公司获得了 AI 融资?

Recent funding coverage on AI Trending includes deals logged in the AI News and Open Source categories. Browse the AI News feed for the latest funding rounds, acquisitions, and valuations. AI Trending 近期收录的融资报道涵盖 AI 资讯和开源项目分类。浏览 AI 资讯 查看最新融资轮、收购和估值信息。

What are the latest AI research breakthroughs? 最近有哪些 AI 研究突破?

The Research section curates the latest papers, model releases, and benchmark results from arXiv, top labs, and industry publications. New entries are added every day. 论文研究 分类精选最新论文、模型发布和基准测试结果,每天更新。

What new AI products launched recently? 最近有哪些新的 AI 产品发布?

Product launches, model releases, and feature updates are tracked in the AI Products category. Coverage includes foundation models, agents, dev tools, and creative tools. 产品发布、模型上线和功能更新见 AI 产品 分类。涵盖基础模型、Agent、开发工具和创意工具。

How is AI regulation changing? AI 监管有哪些新变化?

AI Trending tracks policy, regulation, and safety incidents in the AI Security and AI Overseas categories — executive orders, EU AI Act updates, regional bans, and notable enforcement actions. AI Trending 在 AI 安全AI 出海 分类追踪政策法规、监管动态和安全事件。

Explore More from AI Trending 探索 AI Trending 更多内容