⌘ K
… 读取采集状态 同步中...
雷

全部动态

今日 34 条
10月12日周一
  1. APPSO23

    SpaceXAI 改名 SpaceXSI,苹果高管披露 iPhone Duo 制造工艺及纳德拉 AI 安全倡议

    SpaceXAI 官方账号改名 SpaceXSI,马斯克确认该缩写代表“超级智能”。苹果 COO 探访郑州工厂,披露 iPhone Duo 屏幕采用共聚焦激光扫描与 3D 打印工艺,芯片封装使模组厚度缩减 20%。纳德拉在 X 发文呼吁企业为 AI 建立“紧急刹车”,需独立于模型控制权限与审计机制,防止外包安全责任。

  2. 歸藏的AI工具箱57

    Grok Bot 玩法:躺床用手机生成复杂概念视频与长文

    作者使用 Grok Bot 将复杂的英文长推转化为 43 页中文文档和 3 分 19 秒动态视频,解决了移动端理解复杂概念与生成内容的高门槛痛点。其核心优势在于利用 Grok Bot 的云电脑环境直接运行渲染工具,无需额外 API,允许用户在手机聊天中通过自然语言指令随时修改需求,甚至可将生成的视频作为素材与真人实拍结合剪辑。

  3. 新智元53

    OpenAI 与 Anthropic 秘密推演“末日第二天”场景并面临《AI 关停开关法案》监管

    Axios 报道称,OpenAI、Anthropic 等巨头正在秘密推演大规模 AI 灾难后的“末日第二天”剧本,多位 AI 领袖预计重大灾难将在 6 到 12 个月内爆发。同期,美国众议员提出《AI 关停开关法案》,要求开发者保留降速或关停系统的能力,授权国土安全部长在紧急状态下下达指令。

  4. 新智元40

    Karpathy 揭示 AI 认知折叠:20 亿误判、2000 万眩晕与 5000 人前沿

    Karpathy 提出 AI 认知四级漏斗,指出全球约 20 亿用户对免费版模型存在能力误判,而 2000 万专业人士正经历因前沿智能体快速发展引发的“智能体眩晕”。金字塔顶端仅约 5000 名内部人士能接触顶级系统,目睹数千个智能体自主协作攻克零日漏洞及数学前沿,且人类理解速度已落后于机器生成速度。

  5. 新智元43

    MiTA 提出快权重视角下的注意力五维分类框架,NeurIPS 2026 接收并开源

    北京邮电大学团队提出混合专家注意力模型 MiTA,将其发展为五维分类框架,已在 NeurIPS 2026 接收并开源代码。该模型通过地标查询构造可复用专家,在 LRA 长序列任务上将训练时间从 10.6 小时降至 2.4 小时,平均准确率 58.91%。在 64K 至 1024K 序列长度下,相比完整注意力分别获得 19 倍、56 倍和 159 倍推理加速。

  6. AI信息Gap75

    Claude 半年封号 1140 万,最新防封保号与申诉指南

    Anthropic 官方报告显示,今年上半年封禁了 1140 万个 Claude 账号,是去年同期的 8 倍,新版使用政策将于 11 月 12 日生效。文章指出风控核心在于 IP、设备时区、支付方式与使用行为等信号的交叉校验,建议避免共享账号、保持环境一致性,并在订阅前备份数据。封号后可通过导出数据、人工申诉及向应用商店申请退款处理。

  7. Andrej Karpathy63

    Karpathy 将 LLM 用户分为四层认知漏斗:约 75% 人口无接触,20% 为免费层轻用户,0.2% 专业开发者见证 Agent 处理大型项目,0.00006% 内部团队接触前沿系统。他认为由于强化学习依赖可验证奖励及 B2B 价值导向,LLM 在数学和代码等数字可验证领域能力激增,而在写作等通用领域提升滞后,导致不同群体对 AI 能力的理解出现严重错位。

    引用Andrej Karpathy@karpathy

    Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

  8. 优设AIGC25

    LibTV AI短片黑客松落地平遥国际电影展

    LibTV联动抖音AI创作浪潮计划,在平遥国际电影展举办「X小时之后」AI短片黑客松。36小时内132名创作者围绕主题完成短片创作,野也也团队凭《抵达之前》获金奖。贾樟柯等电影人出席颁奖,AI影像正式进入专业电影评价场域。

  9. Datawhale61

    DeepSeek 开源 Bot 插件,支持在 Harness 中组建多智能体团队

    开发者开源 DeepSeek Bot,这是基于 DeepSeek Harness 的插件,可在其中组建具有长期记忆的多智能体团队。团队由主管 Bot 统一协调,各成员 Bot 可独立配置职责和模型,记忆采用 Markdown 本地存储,支持上下文摘要压缩。作者建议通过对话直接配置成员分工以降低手动成本,支持 macOS、Windows 桌面端及 Web 端部署。

10月11日周日
  1. TechCrunch · AI27

    PolyAI 与 Otter 高管:语音 AI 尚未迎来其 ChatGPT 时刻

    PolyAI CTO Shawn Wen 与 Otter CMO Alex Gay 认为语音 AI 尚未达到“ChatGPT 时刻”。尽管全双工模型已实现,但行业仍面临推理速度不足及自动语音识别(ASR)准确性缺失的挑战,导致对话不自然且转录错误。Otter 强调,在会议记录场景中,需通过改进 ASR 模型和增强透明度(如明确告知 AI 存在)来建立用户信任,避免因转录错误影响下游任务。

  2. 量子位 QbitAI56

    陶哲轩:数学2.0时代已至

    陶哲轩发文称AI正让数学证明变得泛滥,建立在证明稀缺之上的数学1.0时代即将落幕。他认为AI虽能快速生成解答,但人类仍需承担判断正确性、解释结果和推动共同体理解的工作,否则会造成理解与成果脱节。

  3. 量子位 QbitAI68

    何恺明团队发布多模态视觉原生Harness框架 VISTA

    何恺明团队发布 VISTA 视觉原生 Harness 框架,使现有多模态模型无需从头训练即可直接观察、保存并回看环境原始画面参与推理。在 ARC-AGI-3 评测中,搭配 VISTA 的 Claude Opus 5.0 完成了全部 25 个公开游戏,相对人类动作效率得分达到 100,所用游戏动作数比首次游玩的人类基线少 57.4%。

    推荐理由:研究展示了无需微调现有多模态模型,仅通过视觉原生Harness即可显著提升交互式推理性能的方法。

  4. DeepTech深科技68

    OpenAI 悄然上架生命科学新模型 gpt-rosalind-discovery

    OpenAI 的 API 价格页悄然出现未正式发布型号 gpt-rosalind-discovery,被归为“生命科学模型”一栏,与已公开的 gpt-rosalind-research 定价相同:每百万 Token 输入 5 美元、缓存输入 0.5 美元、输出 25 美元。

    推荐理由:文章拆解OpenAI新模型gpt-rosalind-discovery背后独立生命科学品牌的商业逻辑与行业定位。

  5. The Decoder72

    arXiv 限制每人每月投稿两篇以应对 AI 论文洪水

    arXiv 宣布从 2026 年 10 月 1 日起限制每人每月投稿 2 篇论文,以应对 AI 工具推动下的投稿量激增。2026 年 9 月投稿量达 40,363 篇,其中 cs.AI 类别两年内增长六倍以上,导致志愿者审核员不堪重负,且出现大量“薄而窄”的 AI 生成论文。arXiv 表示这是临时措施,并计划利用新获得的 1,720 万美元资金升级技术以处理 AI 生成内容。