即梦正式上线「短剧工作室」；HTC 发口袋 AI 互动伴侣 VIVEE；OpenAI 低成本新模型 GPT-6.1 Terra 发布。
Anthropic 公开递交纳斯达克 IPO 申请：目标 11 月 16 日当周上市
Anthropic 正式踏上上市之路。这家成立仅五年、身处生成式 AI 浪潮核心的企业，本次 IPO 有望创下历史纪录。
Anthropic 在提交给美国证券交易委员会（SEC）的文件中披露，计划登陆纳斯达克交易所，股票代码定为「ANT」，目标在 11 月 16 日当周进行 IPO。该公司将设立 T 类特殊股票，供 T 类董事持有。
在招股书中，该公司披露，上半年营收 162.9 亿美元，上年同期为 11.6 亿美元；净亏损 807.5 亿美元，上年同期为 129.7 亿美元；营业亏损 6.93 亿美元，上年同期为 21.4 亿美元。
截至 9 月 30 日的九个月内，该公司没有任何单一客户占公司营收的 10%以上。79%的收入来自前五大客户以外的客户。合作方面，根据与 SpaceX AI 的协议，该公司同意在截至 2029 年的期限内累计支出高达 1205 亿美元。
该公司透露，资本支出将大幅增加，可能需要比目前预期更早地寻求额外融资；在可预见的未来不会支付任何股息，且目前未就重大收购达成任何协议；可能发行总额最高达 420 亿美元的可转换本票。
截至 9 月 30 日的九个月内，该公司没有任何单一客户占公司营收的 10%以上。79%的收入来自前五大客户以外的客户。合作方面，根据与 SpaceX AI 的协议，该公司同意在截至 2029 年的期限内累计支出高达 1205 亿美元。（来源：新浪科技）
苹果回应 iPhone Duo 内屏开裂问题：仅少量设备受影响，可申请换机
iPhone Duo 上市不久，就有部分用户反馈铰链故障，导致内屏出现裂纹。苹果现已在给 MacRumors 的一份声明中回应了相关消息，称受影响的设备数量很少。
苹果没有说明裂痕产生的原因，也没有透露是否会推出修复方案。受影响的用户可以到线下门店更换新机，但由于备货有限，部分用户需要等待。
苹果表示，该问题波及范围很小。绝大多数用户不会遇到 iPhone Duo 屏幕故障，也不用担忧这个问题会随着使用慢慢出现。出现故障的设备，裂痕大多在购机后不久就会显现。
据了解，一部分网传的裂痕图片是 AI 生成的，这也加大了判断问题真实影响范围的难度。
不少出问题的 iPhone Duo，裂痕位置都集中在内屏下半区，沿着铰链分布。用户称，屏幕损坏前会有卡顿卡滞的手感，有时还会听到咔哒声响。维修拆解网站 iFixit 拆解了 iPhone Duo，找到了几种可能造成铰链故障的原因。该网站表示，铰链这个位置汇集了多个零件，铰链内部一根本该收回的中部支撑杆可能无法正常回缩，进而造成屏幕损伤。（来源：IT 之家）
OpenAI 低成本新模型 GPT-6.1 Terra 深夜炸场：每项任务成本仅 Claude Opus 5.5 九分之一，更快更准确
OpenAI 今天凌晨宣布推出新模型 GPT-6.1 Terra，主打以更低成本完成复杂的电脑操作任务，尤其擅长需要视觉推理和操作桌面软件的工作。与 GPT-5.6 Terra 相比，新模型在事实准确性和编程能力方面也有大幅提升。
Terra 主要面向希望降低自动化成本的企业，可用于处理发票、解读技术图纸等工作。OpenAI 称，面对难度较高的专业任务，Terra 也能以远低于高价模型的成本完成。在考察复杂专业工作的 Agents』Last Exam 基准测试中，Terra 的成绩明显提高。测试项目包括根据建筑图纸建立三维模型，以及逐一记录电影镜头，供影片修复团队使用。
采用中等推理强度时，Terra 的得分比 GPT-5.6 Terra 高 15 个百分点，完成任务的成本却只有后者的约三分之一。与 GPT-6 Astra 相比，Terra 的测试成绩不相上下，每个 Token 的价格却仅为后者的约十分之一。
Terra 在速度上同样具有优势。OpenAI 称，其任务执行速度是 GPT-6 Astra 和 GPT-6.1 Sol 的 3 倍以上。在最高推理强度下，每项任务的成本也仅为 Claude Opus 5.5 的约九分之一。
GPT-6.1 Terra 于 10 月 27 日向Plus、Pro、Enterprise 和 Edu 用户开放，可在 ChatGPT Work 和 Codex 中使用。开发者也能通过 OpenAI API 调用。
API 标准定价方面，每 100 万个输入 Token 收费 1 美元（现汇率约合 6.7 元人民币），每 100 万个缓存输入 Token 收费 0.05 美元（现汇率约合 0.33 元人民币），每 100 万个输出 Token 收费 5 美元（现汇率约合 33.4 元人民币）。OpenAI 称，新模型的价格比 GPT-5.6 Terra 和 GPT-6.1 Sol 低 50%。按照 OpenAI 的准备框架评估，Terra 的网络安全能力已达到「关键级」门槛。（来源：IT 之家）
Meta 宣布至多 950 亿美元重大融资计划，为人工智能等业务筹集资金
10 月 26 日消息，Meta 宣布了总额至多 950 亿美元（注：现汇率约合 6,343.57 亿元人民币）的重大融资计划，为企业核心业务和新兴人工智能领域的持续发展提供资金支持。
这 950 亿美元分为多个部分，包括 225 亿美元的 A 类普通股远期出售、125 亿美元的强制可转换优先股、后续至多 600 亿美元的潜在 C 类普通股按市价发行。NVIDIA（英伟达）、AMD（超威）、Broadcom（博通）均有意购买至多 50 亿美元的 Meta 的 A 类普通股。
Meta 在新闻稿中表示，其在人工智能方面的投资已通过提升用户参与度、改善广告效果、提高盈利能力等方式产生了切实的回报。
该企业预计 2026 年资本支出达 1,350~1,450 亿美元，Meta 计划通过运营现金流、投资级债务、战略合作、股权资本的均衡组合来为人工智能投资提供资金。（来源：新浪科技）
消息称腾讯绿泡泡短剧小程序月活过亿，10 月首周周活环比增长 44.1%
QuestMobile 数据显示，截至今年 8 月，腾讯绿泡泡短剧小程序月活跃用户规模已过亿级，达 1.43 亿。10 月首周周活跃用户达 8968 万，较 9 月首周提升 44.1%。自 7 月正式更名以来，该产品周活跃用户规模约三个月内逼近 9000 万。
DataEye 监测数据显示，10 月 1 日至 7 日绿泡泡短剧累计投放素材 4.3 万组，日均投放量较节前一周增长 57%，内容供给与投放投入同步加码。
从增长驱动来看，原「短剧」小程序入口的全面接替，与微信生态流量支持为绿泡泡带来直接增量。依托微信生态的分发，拥有 6 亿日活用户的视频号是其中关键一环；另一方面，绿泡泡持续扩充内容供给，覆盖真人剧、AI 剧、动漫剧等品类。
「绿泡泡短剧」是腾讯开发的微信小程序。该产品界面分为「免费剧场」「刷一刷」和「我的」三部分。「免费剧场」下设「AI 剧」「真人剧」「每日上新」「动漫剧」「本周必看」「百万播放」等栏目。点击「刷一刷」入口，上下滑动即进入沉浸式推荐流。剧集全部免费，但会在剧集之间插播广告。（来源：IT 之家）
腾讯混元同声传译大模型 Hy-SimulMT 开源，响应时间仅 3 秒超人类译员
10 月 26 日消息，腾讯混元同声传译大模型 Hy-SimulMT 今日发布并开源，具备低时延、高准确率的特点，支持中英语言之间语音到语音实时翻译， 说话人的音频可以实时输入到 Hy-SimulMT 模型中，模型会边听边译，实时给出译文和语音，并支持实时音色克隆，可以高度还原说话人声音和语气。该模型在腾讯 Hy 翻译 App 首发，免费提供服务体验。
低时延是模型的一大亮点，混元同声传译大模型 Hy-SimulMT 的响应时间实测在 3 秒左右，人类专业译员平均水平在 4-5 秒。测评结果显示，该模型的翻译效果超过了市面上主流的开源同传模型，甚至超过大部分的商用同传 API。
官方称，在业界常用的长音频开源评测集 RealSI 上，Hy-SimulMT 展示出了较好的同传翻译效果。为了更准确评估 Hy-SimulMT 的真实水平，该团队在合规允许范围内，广泛收集了互联网上新出现的中文和英文音频，包含软件 AI、医学、制造能源、自然科学、农业食品、商务管理、法律、教育、文化文旅、体育生活共十个领域，涉及演讲、访谈、会议等多个场景。在这份数据上，该团队不仅测试了 Hy-SimulMT 的效果，也请多位人类同传专家对这些音频进行同声传译。
人类同传专家一共 6 位，其中 3 位拥有 CATTI 一级口译证书，3 位拥有 CATTI 二级口译证书，所有专家都有多年的同传实战经验。测评对 Hy-SimulMT 和人类同传专家的翻译内容在延迟和翻译质量上对比，其中人类同传专家选取的是 3 个人中的最高分，在这次能力定位测评中，Hy-SimulMT 在翻译的质量达到人类同传专家水平（得分更高），速度更快（时延更低）。（来源：IT 之家）
整合小云雀成熟能力 即梦正式上线「短剧工作室」
10 月 26 日消息，即梦 AI 上线全新功能「短剧工作室」。作为即梦面向短剧创作场景推出的一站式 AI 工具，「短剧工作室」将剧本整理、角色与场景管理、分镜生成、视频制作及后期调整等全流程整合到同一个平台中，让创作者不必在多个平台间来回切换。
今年 3 月，与即梦同属剪映旗下的内容创作平台小云雀率先推出短剧 Agent。据悉，「短剧工作室」部分功能整合了小云雀短剧 Agent 的同款能力，并在此基础上进行了更新升级。
即梦与小云雀此前在工具形式上各自有一些探索，都积累了不错的口碑，尤其是即梦对于 eedance 系列模型的能力适配，一直比较领先，此次能力整合能够让创作者在一个平台里找到更加丰富的工具形态，可以更好地发挥各自优势。
即梦相关负责人表示，内容创作的形式在不断变化，但创作者对高效、统一的创作工具的需求是共通的。即梦希望持续把经过验证的 AI 创作能力整合到一个平台上，让不同类型的创作者都能在熟悉的工具里完成从灵感到成片的完整流程，把更多精力放在创意本身。（来源：新浪科技）
1 秒下载两万部 4K 电影：3.2Pb/s，我国团队创下单纤双向实时光传输容量的世界纪录
10 月 26 日消息，据科技日报从中国信科集团光通信技术和网络全国重点实验室获悉，该实验室日前联合鹏城实验室、烽火藤仓光纤科技有限公司，成功完成 3.2Pb/s（拍比特每秒）多芯光纤实时双向传输试验，创下单纤双向实时光传输容量的世界纪录。
「3.2Pb/s 的传输速率，意味着可在一秒内下载两万余部 4K 电影。」中国信科集团光通信技术和网络全国重点实验室光系统高级工程师杨超表示，该方案验证了多芯光纤超宽带扩容路线产业化潜力，为大型算力枢纽互联、骨干传输及 6G 承载提供下一代高速光通信核心技术支撑。
随着人工智能、大模型计算及数据中心规模化发展，网络流量呈爆发式增长，传统单模光纤传输系统逐步逼近容量极限。针对这一瓶颈，团队创新提出 24 芯多芯光纤双向传输架构，采用 10.3 公里 24 芯多芯光纤，融合 S+C+L 三波段复用与空分复用技术，构建数千路并行传输通道，使总带宽规模大幅突破传统单模光纤容量上限。系统搭载商用相干光模块，采用高阶概率整形调制，实现稳定实时双向信号传输；通过芯层排布优化有效抑制芯间串扰，简化信号均衡处理，兼顾传输性能与工程实用性。
「相比传统 C+L 波段商用单芯 G.652 光纤约 100Tb/s 的容量上限，该系统将单纤总传输容量提升 32 倍；完成 S 波段单波 400G、C+L 波段单波 1.2Tb/s（太比特每秒）自研光模块的实时传输验证，突破了仅依靠光谱扩容难以逾越的单芯光纤容量墙。」杨超介绍。目前本次成果已作为邀稿投至国际光学权威期刊《Journal of Lightwave Technology》。（来源：IT 之家）
苹果推出 699 美元新款 iPad Mini 配备 OLED 屏和防水设计
苹果公司周一发布改版 iPad mini，配备 OLED 显示屏、防水设计和速度更快的芯片，由此拉开新品发布周的序幕。
这款平板电脑定价 699 美元，配备 8.5 英寸 OLED 显示屏，可呈现更深的黑色、更亮的白色和更高的对比度。苹果通过缩窄屏幕边框，将显示屏尺寸从之前的 8.3 英寸扩大至 8.5 英寸。改版机型还采用新的「Acoustic Display」音响系统，扬声器置于显示屏之下，使其成为首款达到 IP67 防尘防水等级的 iPad。
新款 iPad mini 比上一代薄 14%、轻 13%。苹果还调整了前置摄像头的位置，因此更适合 FaceTime 通话和视频会议。
新机搭载 A20 Pro 芯片，较上一代采用的 A17 Pro 处理器升级；上一代起售价为 599 美元。新款提供标准显示屏和哑光纳米纹理显示屏选项，存储容量从 128GB 到 512GB。配备最高存储容量、蚀刻哑光屏幕和蜂窝网络连接的顶配版售价为 1249 美元。
这一新款 iPad mini 的发布拉开了苹果 10 月底系列新品发布的序幕。本周晚些时候，公司计划发布搭载 M6 芯片的新款 iMac 和 14 英寸 MacBook Pro，并将在本周结束前推出首款触摸屏 MacBook Pro。（来源：新浪科技）
HTC 公布口袋 AI 互动伴侣 VIVEE：支持个性化性格，预售价 99.99 美元
10 月 26 日消息，HTC（宏达电子）近日公布了一款名为 VIVEE 的口袋 AI 互动伴侣，其预售价为 99.99 美元（现汇率约合 667.7 元人民币）。
用户抚摸、拍打、转动 VIVEE 时，该设备均会给出有趣互动回应；其也支持聊天、玩游戏，在与用户的互动中，这个口袋伴侣会逐渐养成自己的小个性。
HTC 还提供 VIVEE Suits 造型配件，在为 VIVEE 提供新造型的同时也能解锁不同的屏幕 UI 表情。
硬件层面，HTC VIVEE 三维 60×60×50 (mm)，质量 114g，配备 OLED 屏幕，具备全天候续航，支持 Wi-Fi 蓝牙、NFC 连接，具备震动反馈，提供 IPX4 防水。该设备具备麦克风与扬声器硬件开关，HTC 承诺用户个人信息不会被用于 AI 模型训练。（来源：IT 之家）
小米开源硬件 HachimoDock 亮相，150 元就能复刻万能 AI 桌宠
小米技术委创新产品团队孵化的开源 AI 硬件 HachimoDock 亮相。这是一台可由用户按个人喜好自由改造的桌面 AI 设备，小巧机身集成了个性化桌宠、语音操控电脑端 Agent、智能家居联动、自定义组件创作等能力。
据介绍，HachimoDock 硬件自制成本约 150 元。团队希望以此降低 AI 硬件的体验门槛，让它成为大众入门 Vibe Coding 的第一台 AI 硬件，体验用自然语言去「创造」一台万能桌面搭子的乐趣。
这台设备的功能不止于此：用户说出想法，即可与 AI 协作，现场生成并运行新的组件。基于这一开放机制，它既可以是快捷语音输入的麦克风、内置万款游戏的小游戏机、收听 FM 电台的小收音机，也可以是查看 AI 任务进展与电脑状态的桌面副屏、联动智能家居的桌面中控……更多玩法，留给每位用户亲手创造。
用户可以通过语音直接与桌宠进行互动，它会将用户意图转化为各类任务派发给电脑端 Codex、WorkBuddy 等各类 Agent，例如查资料、整理文件、编写代码等，同时 HachimoDock 屏幕上也会同步展示思考、执行、等待确认等任务状态，并以桌宠的表情、动作、简短提示及声音提醒反馈进展。
HachimoDock 的语音操作还延伸到了家居场景。通过联动小米 Miloco，用户可以语音发起家居控制，小屏也可切换为家居控制台，呈现设备状态、灯光控制和监控画面等内容，让一台桌面设备同时成为电脑 Agent 和智能家居的交互入口。目前，HachimoDock 已在 GitHub 公开软件源码，并提供硬件物料清单、结构件模型及装配教程，用户可以按照自己的想法随意修改。（来源：IT 之家）
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Jimeng officially goes online as a "short play studio"; HTC handbag AI interactive partner VIVEE; Open AI new low-cost model GPT-6.1 Terra publish.
Anthropic publicly submitted NASDAQ IPO application: Target for listing on November 16th
Anthropic is officially on the market. This is a five-year-old company at the heart of the generation of AI waves, and this time IPO is expected to be a record.
Anthropic disclosed in a paper submitted to the United States Securities and Exchange Commission (SEC) that it planned to land on the NASDAQ Exchange with the stock code known as "ANT" and that the target was an IPO on November 16th. The company will set up special T-type shares for T-class directors.
In its equity book, the company disclosed that it had received $16.29 billion for the first half of the year, compared with $1.16 billion for the same period of the previous year; a net loss of $80.75 billion, compared with $12.37 billion for the same period of the previous year; and an operating deficit of $693 million, compared with $214.0 billion for the same period of the previous year.
In the nine-month period ending 30 September, no single customer of the company accounted for more than 10 per cent of the company's receipts. Seventy-nine per cent of the income came from clients other than the top five. In terms of cooperation, under the agreement with SpaceX AI, the company agreed to aggregate expenditure of up to $12.5 billion for the period up to 2029.
The company disclosed that capital expenditure would increase significantly and that additional financing might need to be sought earlier than currently anticipated; that no dividends would be paid in the foreseeable future and no agreement had been reached on major acquisitions; and that a total of $42 billion in convertible promissory notes might be issued.
In the nine-month period ending 30 September, no single customer of the company accounted for more than 10 per cent of the company's receipts. Seventy-nine per cent of the income came from clients other than the top five. In terms of cooperation, under the agreement with SpaceX AI, the company agreed to aggregate expenditure of up to $12.5 billion for the period up to 2029. (Source: New Wave Technology)
Apple response to iPhone Duo internal screen breakup: Only a small number of equipment is affected and can apply for replacement
Shortly after the iPhone Duo came on the market, some users responded to a chain failure that led to a crack in the inner screen. Apple has now responded to the information in a statement to MacRumors that the number of equipment affected is very small.
The apples did not explain the cause of the cracks, nor did they indicate whether a rehabilitation programme would be introduced. The affected users can go to the lower-line store to replace the new machines, but some of them have to wait because of limited availability.
Apple indicates that the problem is very small. The vast majority of users do not experience an iPhone Duo screen failure, nor do they have to worry that the problem will slowly arise with use. Most of the cracks in the malfunctioning equipment will appear shortly after the machine is purchased.
It is understood that some of the web-borne fissure images are created by AI, which makes it more difficult to judge the real extent of the problem.
There are a number of problems with iPhone Duo, where cracks are concentrated in the lower half of the inner screen and distributed along the chain. Users claimed that the cartons were held back by hand before the screen was damaged, and sometimes the sound of the sound was heard. The iFixit site was repaired and the iPhone Duo was dismantled, and several possible causes of chain failure were identified. The site indicates that this position of the chain brings together several parts and that a central support pole within the chain that is to be recovered may not be able to retort normally, causing damage to the screen. (Source: IT Home)
Open AI New Low Cost Model GPT-6.1 Terra Late Night Bombing Field: Only Claude Opus 5.5 a ninth of the cost per assignment, faster and more accurate
Open AI announced this morning the launch of the new model GPT-6.1 Terra, which focuses on more low-cost complex computer operations, especially those that require visual reasoning and desktop software. Compared to GPT-5.6 Terra, the new model has also significantly improved its factual accuracy and programming capacity.
Terra is primarily for businesses that wish to reduce the cost of automation and can be used to process invoices, interpret technical drawings, etc. Open AI states that in the face of more difficult professional tasks, Terra can also do so at a cost far below the high-price model. Terra's performance improved significantly in the Las Exam benchmark test, which examined complex professional work. The test projects included the development of a three-dimensional model based on building drawings and the recording of the film shot on a case-by-case basis for use by the film repair team.
When medium reasoning was used, Terra scored 15 percentage points higher than GPT-5.6 Terra, while the cost of completing the mission was about one third. In comparison to GPT-6 Astra, Terra's test scores were very high, and each Token price was only about one tenth of the latter.
Terra has the same advantage in speed. Open AI claims that the speed of execution of its mission is more than three times that of GPT-6 Astra and GPT-6.1 Sol. At the highest level of reasoning, the cost of each task is only about one ninth of Claude Opus 5.5.
GT-6.1 Terra was made available to users in Plus, Pro, Enterprise and Edu on October 27, and can be used in ChatGPTWork and Codex. Developer can also call through Open AI API.
For API standard pricing, Token charges US$ 1 per 1 million input (current exchange rate approximately RMB 6.7), US$ 0.05 per cache input price (current exchange rate approximately RMB 0.33), and US$ 5 per 1 million output (current exchange rate approximately RMB 33.4). Open AI states that the price of the new model is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. According to Open AI's preparedness framework assessment, Terra's network security capability has reached the critical level threshold. (Source: IT Home)
Meta announced a major financing plan of up to $95 billion to raise funds for business such as artificial intelligence
On October 26, Meta announced a major financing plan totalling up to $95 billion (note: the current exchange rate is about $6,34357 million), which provides financial support for the continued development of core business and emerging areas of artificial intelligence.
This $95 billion is divided into several components, including $22.5 billion in forward sales of general A-type shares, $12.5 billion in mandatory transferable priority shares, and up to $60 billion in follow-up to potential general C-type shares. NVIDIA (In Weeda), AMD (Upper Power) and Broadcom (Book) are interested in buying up to $5 billion in Meta's general A shares.
Meta stated in the press release that its investment in artificial intelligence had yielded tangible returns by increasing user participation, improving advertising effectiveness and increasing profitability.
The enterprise expects capital expenditure to reach $1,350 to $145 billion in 2026, and Meta plans to finance artificial intelligence investment by operating a balanced mix of cash flow, investment-level debt, strategic cooperation and equity capital. (Source: New Wave Technology)
According to the news, the Tencent Green Bubble Bubble Mini-Program survives hundreds of millions a month, with the first week of October growing by 44.1 per cent.
QuestMobile data show that, as of August of this year, the Tencent Green Bubble Mini-Plus monthly active users were over 100 million degrees, reaching 143 million. The number of active users in the first week of October was 89,688,000, up 44.1 per cent from the first week of September. Since the official change of name in July, the number of active users in the product week has approached 9 million within approximately three months.
Data from DataEye monitoring shows that a total of asset 43 million green bubbles were dropped from 1 to 7 October, with a 57 per cent increase in average daily delivery compared to the week before the festival, and a synchronized addition of content supply and input.
From a growth-driven perspective, the full-scale replacement of the former "short play" applet portal with the WeChat ecological flow support brought a direct increase in green bubbles. A key link is the video number of 600 million live users, based on the distribution of the WeChat ecology; on the other hand, the green bubble continues to expand the content supply to cover such items as reality dramas, AI dramas and cartoons.
"Green Bubble Shorts" is a WeChat applet developed by Tencent. The product interface is divided into three parts, "Swipe a brush" and "My". The "Ai" "The real play" under "The free theater" "The new daily" "The cartoon" "Absolutely this week" "Million play" etc. Click at the "Swipe" entrance and move up and into the immersion referral stream. The series is free of charge, but advertisements are broadcast between the series. (Source: IT Home)
Open source LLM Hy-SimulMT, response time 3 seconds superhuman interpreter
LLM Hy-SimulMT Today, publish, with low latency and high accuracy, supports live voice-to-voice translation between Chinese and English languages, and the speaker ' s voice can be entered in real time into the Hy-SimulMT model, where the model is heard and translated, translated and spoken in real time, and supports real-time voice cloning, so that the voice and tone of the speaker can be restored at a high level. The model was launched in Tencent Hy Translator App, free of charge.
Low latency is one of the bright spots of the model, and the response time for a hybrid simultaneous LLM Hy-SimulMT is measured at around 3 seconds, with an average human professional interpreter at 4-5 seconds. The results showed that the model was translated more effectively than the main open-source simultaneous model in the market, or even more than most commercial simultaneous API.
Officially, Hy-SimulMT shows a better co-translation effect on RealSI, which is commonly used in industry. In order to assess more accurately the true level of Hy-SimulMT, the team, within the limits of compliance, collected a wide range of new and emerging Chinese and English audio on the Internet, including software, AI, medicine, manufacturing energy, natural sciences, agro-food, business management, law, education, cultural and cultural tours, sports life and so on. In this data, the team not only tested the effects of Hy-SimulMT, but also requested simultaneous interpretation of the audio by a number of human simultaneouss.
The total number of experts in the same field is six, of whom three have CATTI Level I interpretation certificates and three have CATTI Level II interpretation certificates, all of whom have experience in the same field for many years. The quality of translations of Hy-SimulMT and human compatriot experts is compared with the quality of translations, where human compatriot experts select the highest score among three individuals, and in this competency positioning test, Hy-SimulMT reaches the level of human compatriot experts (higher scores) and is faster (latency is lower). (Source: IT Home)
Jimeng officially goes to the Shorts Studio.
On October 26, Jimeng AI went online with a brand-new feature, the "short play studio". As a all-in-one AI tool for short play creation scenes, the "short play studio" integrates the whole process of script preparation, role and scene management, storyboard generation, video production and post-production adaptation into the same platform, so that creators do not have to switch back and forth between multiple platforms.
In March of this year, the Little Skylark, which is part of Jimeng's CapCut-based content creation platform, spearheaded the launch of the short play Agent. According to the information received, part of the short play studio function integrates the same-term capacity of Agent, a small sparrow, and has been upgraded on that basis.
Jimeng and Little Skylark have had some previous explorations of the form of the tool, and have accumulated a good reputation, especially as Jimeng has been relatively ahead of the edance model, and this integration of capacity allows creators to find a more rich form of tool in a platform and to better exploit their respective strengths.
The person responsible for Jimeng said that the form of content creation was changing, but that there was a common demand among creators for efficient and unified creative tools. Jimeng hopes to continue to integrate the proven AI creative abilities into a platform where different types of creators can complete a whole process of spiritualization in a familiar tool and focus more on creativity itself. (Source: New Wave Technology)
1 second download of 20,000 4K films: 3.2Pb/s, a world record of one-sided, two-way real-time light transmission capacity
On 26 October, according to a news report from the Daily Science and Technology newspaper, the China Sinko Group Light Communication Technology and Network National Focus Laboratory, was able to successfully complete the 3.2 Pb/s (bits per second) real-time two-way multi-channel fibre transfer experiment with a world record of one-sided real-time light transmission capacity at the former U.P.U.U.U.U.U.U.U.U.U.U.U.U.U.U.U.U.
"3.2 Pb/s' transmission rate means that more than 20,000 4K films can be downloaded in one second. According to Yang, the senior engineer of the China Shinko Group Light Communication Technology and Network National Focus Laboratory Light Systems, the program validated the industrial potential of multichip fibre-optic ultrabroadband expansion routes, providing the next generation of high-speed light communications core technology support for large computing hub interconnection, backbone transmission and 6G load.
With the development of artificial intelligence, LLM computing and the scaling up of data centres, there has been an explosive growth in network traffic, and the traditional single-modulator fibre-optic transmission system has gradually reached its capacity limit. In response to this bottleneck, the Team Innovation proposes a 24-core multi-chip fibre-optic two-way transmission structure that integrates the S+C+L three-band reuse and space-sharing technology with 10.3 km of core-chip fibre-optics, and builds thousands of parallel transmission channels that will allow the total bandwidth to significantly exceed the traditional single-module capacity cap. The system carries a commercial symmetric light module, which uses a high-probability integrator system to achieve a stable real-time two-way signal transfer; optimizes core-scraping to effectively contain inter-chip-strength, simplifys signal balance processing and balances transmission performance with engineering utility.
"The system increases the total fibre transfer capacity by 32 times compared to the capacity limit of the traditional C+L-band commercial single core G.652 fibres of about 100 Tb/s; completes the real-time transmission validation of the S-band single-wave 400 G, C+L-band single-wave 1.2Tb/s (Tabit per second) self-research modules, breaking through the wall of single-chip fibre-optic capacity, which cannot be exceeded by spectral expansion alone. I'm sorry. The results have now been sent as an invitation to the International Optical Authority Journal Journal of Lightwave Technology. (Source: IT Home)
Apple launch $699 new iPad Mini with OLD screen and waterproof design
Apple is re-enacting the iPad Mini on Monday with an OLED display screen, waterproof design and faster chip, thus opening the new publish week.
The tablet is priced at $699 and equipped with 8.5 inches of OLD display screens, showing a deeper black, brighter white and higher contrast. The apple expands the size of the display screen from the previous 8.3 inches to 8.5 inches by narrowing the screen border. The modified machine also uses the new Acoustic Display sound system, with speakers placed below the display screen, making it the first iPad of the IP67 dustproof water-preserve level.
The new iPad Mini is 14% less than the previous generation. Apple has also repositioned the front camera and is therefore more suitable for FaceTime calls and videoconferences.
The new aircraft carries the A20 Pro chip, which is upgraded from the A17 Pro processor used by the previous generation, which started at $599. The new paragraph provides the standard display screen and dumb-ray nanotextile screen options with storage capacity ranging from 128 to 512 GB. The top-end sale price with the maximum storage capacity, the erosion of the mute screen and the cellular network is $1249.
The new iPad Mini's publish pulls the apples and opens the end of October with a new series of publishs. Later this week, the company plans publish to carry the new iMac and 14 in MacBook Pro, and will launch the first touch screen MacBook Pro by the end of the week. (Source: New Wave Technology)
HTC Publication Pocket AI Interactive Partner VIVEE: Supporting Personality, Presale Price 99.99
On October 26th, HTC (Hunda Electronics) recently published a pocket AI interactive partner named VIVEE, with a pre-sale price of US$ 99.99 (current exchange rate of approximately RMB 667.7).
The device provides interesting interactive responses when the user caressssssssss, beats, turns VIVEE; it also supports chats, plays games, and in interaction with the user, the pocket partner develops his or her own small personality.
HTC also provides VIVEE Suits configurations that can unlock different screen UI expressions while providing new configurations for VIVEE.
Hardware level, HTC VIVEE 3D 60x60x50 (mm), mass 114g, equipped with OLED screens, full-time re-routing, supporting Wi-Fi Bluetooth, NFC connection, generating vibration feedback, providing IPX4 water protection. The device has a microphone and speaker hardware switch, and HTC promises that user personal information will not be used in AI model training. (Source: IT Home)
The millimetre open-source hardware, HachimoDock.
Open source AI hardware HachimoDock for MiTechnology team hatching innovation products. This is a desktop AI device that can be freely adapted by the user to his/her preferences, and the small pilot body brings together personalized table pets, voice manipulations of the computer, Agent, smart-house-connection, self-defined component creation, etc.
It was described that HachimoDock hardware had a cost of approximately $150. This is how the team wants to lower the threshold of the experience of the AI hardware, making it the first of its kind to be introduced to the public, and to experience the pleasure of "creating" an all-powerful desktop set in a natural language.
This device has more than this function: users can work with AI to generate and run new components on-site. Based on this open-ended mechanism, it can be either a microphone entered by a fast voice, a small game player with a million-dollar built-in game, a small radio listening to FM radio, or a desktop subscreen to view AI mission progress with computer status, a desktop control to connect smart homes... More games, left to each user.
Users can interact directly with table pets by voice, which translates the user ' s intentions into various tasks sent to computer users, such as Cordex, WorkBuddy, and Agents, such as checking information, organizing documents, writing codes, etc., while the HachimoDock screen simultaneously displays the status of the task, such as thinking, execution, waiting for confirmation, and alerts the progress of the feedback with the table pet expression, action, short tips and sound.
HachimoDock's voice also extends to home scenes. By connecting Miloco, the user can initiate home control by voice or switch the screen to a home control table, displaying the state of the equipment, light control and surveillance, and making a desktop device an interface between the computer Agent and the intelligent home. At present, HachimoDock is in GitHub open software source code and provides a list of hardware materials, a model of structural components and a assembly curriculum, which users can modify at their own discretion. (Source: IT Home)
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.