OpenAI 宣布已通知 SpaceX,计划终止向 Cursor 提供 OpenAI 模型的合作,拟定的停止服务日期为 2026 年 11 月 12 日。 按照现有合同,OpenAI 选择了允许范围内最长的通知期,希望尽可能延长 Cursor 用户继续使用这些模型的时间。 这项决定发生在 Cursor 被 SpaceX 收购之后。OpenAI 表示,与 SpaceX 这样的大型合作伙伴合作时,通常会通过定制合同约束模型使用方式,并确保大规模集成符合安全要求。但基于过去与 Elon Musk 旗下公司的合作经历,OpenAI 认为自己无法确信 SpaceX 会持续遵守相关服务条款。 OpenAI 在声明中提到,Musk 收购 Twitter 后,Twitter 曾违反双方合同。今年早些时候,Musk 也在宣誓作证时承认,xAI 曾违反 OpenAI 的服务条款。 OpenAI 与 Cursor 的协议中包含一项控制权变更条…

Channel
Levix 空间站
@synctoai
On this record: Growth · Engagement · Reactions · Posts · Telegram's recommendations · Cite this entry
1,105subscribers
+4 since we began measuring on 29 August 2026
Risers and fallers across the register · movement among entries of 1,000–3,162.
Register entry
| Telegram ID | -1001768561381 |
|---|---|
| Type | Channel |
| Username | @synctoai |
| Created | Between 1 December 2021 and 30 April 2023 — estimated from Telegram’s id allocation, not measured. How this range is calculated. |
| First recorded | 29 August 2026 |
| Last confirmed live | 11 September 2026 |
| Measurements held | 4 |
| Confirmed unchanged | 2 times, most recently 11 September 2026 |
| On Telegram | t.me/synctoai |
Growth
| Measured (UTC) | Subscribers | Change |
|---|---|---|
| 6 Sept 2026, 19:56 | 1,105 | +1 |
| 2 Sept 2026, 12:28 | 1,104 | +3 |
| 30 Aug 2026, 17:17 | 1,101 | no change |
| 29 Aug 2026, 22:30 | 1,101 | first reading |
Engagement
20 posts held, back to 24 August 2026 — the reader has not yet reached the start of this channel’s public history, so older posts may sit further back, unread. Read across 1 page of Telegram’s post history, 20 posts per page.
- ERR · 30 days
- 3.07%
- avg views ÷ 1,105 subscribers
- Avg views / post
- 33.9
- 20 posts measured
- Reaction rate
- 2.44%
- reactions ÷ views · ER floor
- Posts in window
- 20
- of 20 held
ERR is average views per post over the last 30 days divided by subscribers, the definition TGStat uses, so this figure is comparable with the one you will see elsewhere. It falls structurally as a channel grows: a high ERR on a small channel and a low one on a large channel describe reach mathematics, not quality. We publish the figure and the sample it came from and pass no verdict on it.
ER is defined industry-wide as (forwards + reactions + comments) ÷ views — note the denominator is views, not subscribers. Telegram’s public web preview carries views and reactions but not forward or comment counts, so the reaction rate above is the reactions term only and is therefore a floor: the true ER for this channel is higher by an amount we have not measured and will not estimate. It is computed over the 1 of 20 measured posts that carry a reaction reading, and over those same posts' views.
| Window | Rolling 30 days · latest post in window 29 August 2026 |
|---|---|
| Posts held | 20 (24 August 2026 – 29 August 2026) |
| Views total | 678 |
| Reactions total | 1 |
| Forwards / comments | not exposed by the public surface — not measured, not estimated |
| Readings taken | 29 Aug 2026, 22:30 UTC |
Views are a single reading per post, taken at the time above. A post published in the last day or two is still accumulating views, which pulls the 30-day average down slightly. That is a property of the standard definition rather than a fault in it, so we keep the definition rather than “correcting” the number into something nobody can reproduce.
Precision. Telegram publishes view counts on its public widget in short form — 8.12K, 3.7M — so any reading at or above 1,000 reaches us rounded to three significant figures, and only counts below 1,000 are exact. Averages and rates derived from them are shown to the same precision rather than to the unit: a figure like 3,701,250 would assert digits nobody measured.
Reaction counts are published per emoji and rounded the same way, so a total below 1,000 is exact and a larger one is a sum that may carry a rounded component from each emoji above 1,000. Because it is a sum, it does not look rounded — read a large reaction total as three significant figures per contributing emoji rather than as the figure it prints.
Reaction mix
1 reaction across 1 post, in 1 kind.
| Reaction | Count | Share | Share, drawn |
|---|---|---|---|
| ❤ | 1 | 100.0% |
No sentiment is inferred, and none should be read in. This table is ordered by count and by nothing else. Emoji do not carry stable meaning across languages or communities — 🙏 is thanks in one channel and mourning in another — so we publish which ones were pressed and how often, and pass no judgement on what an audience meant by them.
Precision. Telegram publishes reaction counts per emoji and short-forms each one — 4.34K, 1.2M — so any single kind at or above 1,000 reaches us at three significant figures, and only counts below 1,000 are exact. The shares above are ratios of those figures and carry the same error. This is also why the total here can differ slightly from a reaction total printed elsewhere on the page: both are sums of the same rounded parts, taken over samples with different edges.
Coverage. Reactions were read on 1 of the 20 sampled posts in this sample. Summed by Telegram’s own count on each post — not by adding up the per-emoji breakdown above — those same posts carry 1 reactions in total: the kind of figure the paragraph above means by “a reaction total printed elsewhere on the page”.
Measured over the 20 most recent posts we hold, published 24 August 2026 to 29 August 2026, using the newest reading held for each. Telegram Stars are excluded: they are a payment, not a reaction, and they have their own section.
Recent posts
Pollen Robotics 推出了一款很特别的小型双足机器人 Microduck。它只有 25 cm 高、约 800 g 重,定位并不是单纯的遥控玩具,而是一台可以自己训练动作的开源机器人。Microduck 已于 2026 年 8 月 27 日开放预订,首发价为 399 美元,不含税费和运费,官方计划在 2026 年圣诞节前发货。 Microduck 到手就能玩,机身内置 15 个电机,并配有摄像头、LiDAR 和两组 IMU,机器人本地可以以 50 Hz 运行策略。包装中包含机器人、电池、USB-C 线和游戏手柄,共提供 7 个已经训练好的动作策略,包括行走、坐下再站起、踢球、低头抓取、穿上滚轮后的滑行,以及摔倒后自己重新站起来。 更有意思的是,这些动作并没有被封装成不可修改的固定功能。Microduck 的每一种行为都是可以重新训练的 policy。用户可以先在物理模拟器里训练,通过本地电脑或 Hugging Fa…
❤1
Cloudflare 如何靠优化 DNS 缓存省下 100 TB 内存 Cloudflare 工程师 Sebastiaan Neuteboom 介绍了 1.1.1.1 背后的 DNS 平台 Big Pineapple 最近一次内存优化。这个系统同时支撑 1.1.1.1、Gateway DNS、DNS Firewall 等服务,任何时刻都保存着超过 2500 亿条 DNS 缓存记录。到了这个规模,每条记录哪怕只浪费 1 个字节,整个集群就会多占用 250 GB 以上内存。 团队从缓存记录在 Rust 中的实际内存布局入手。第一处调整,是把写入缓存后不会继续增长的 Vec<T> 和 String 换成 Box<[T]> 与 Box<str>。这样可以去掉无用的 capacity 字段和预留空间。每条缓存里有 8 个类似字段,仅这一项就能减少 64 字节,放大到整个集群,相当于节省超过 15 TB。 接下来,他们把 DNS 响应…
OpenExecutive 是 SenteLabsAI 开源的一套 AI 虚拟高管系统。使用者面对的始终是一个统一的“Executive”,但在后台,它可以调度 8 个不同领域的智能体,分别负责战略、财务、人力、法务、运营、市场、产品和董事会沟通。遇到复杂问题时,系统会把任务分给相应角色并行处理,再把结果整合成一份口径一致的管理建议。 它并不只是把几个 Claude Agent 放在一起。OpenExecutive 内置了一套 MBA 级知识库,同时允许企业上传商业计划、财务模型、战略文件等内部资料。每个专业智能体在回答问题时,都可以通过 ChromaDB 检索公共知识和企业自己的文档。首次使用还会通过向导收集公司所属行业、发展阶段、商业模式、竞争环境、战略重点、文化等信息,让后续回答尽量建立在具体公司背景上。 系统还加入了长期使用需要的能力。每轮对话结束后,会由较轻量的模型提取重要决策、项目和建议,保存到 SQLite 中…
Nvidia 正在与 Hugging Face 讨论一笔可能超过 130 亿美元的收购交易。知情人士向 Business Insider 表示,双方最近几周一直在接触,但截至 8 月 27 日尚未签署协议,谈判仍有可能终止。Nvidia 和 Hugging Face 都没有对此发表评论。 这笔交易如果落地,将成为 Nvidia 规模最大的收购之一。随着 AI 业务持续扩张,Nvidia 也在明显加快对外投资。公司最新披露,本财年剩余时间已经承诺投入约 180 亿美元进行股权投资,此前持有的私人公司投资规模已经达到 479 亿美元。Microsoft 也曾与 Hugging Face 接触,不过相关谈判目前已经停止。 Nvidia 与 Hugging Face 并不陌生。2023 年,Nvidia 参加了 Hugging Face 的 2.35 亿美元融资,当时公司的估值为 45 亿美元。2025 年底,Nvidia 又提出投…
OpenAI 自研推理芯片 Jalapeño,已经跑到什么水平? SemiAnalysis 近日披露了 OpenAI 自研推理芯片 Jalapeño 的更多细节。OpenAI 从 2024 年中开始组建团队并推进设计,约 16 个月后完成流片。不同于外界常见的理解,这颗芯片并非只针对 OpenAI 自家模型,而是一颗面向大模型推理的通用 ASIC,DeepSeek R1、Kimi K2.5 和 GPT-OSS 都已经能够运行。 目前最亮眼的是能效。SemiAnalysis 在 OpenAI 实验室验证的 InferenceX 测试中,Jalapeño 在多数场景下的每瓦性能超过 Nvidia Blackwell;即使拿来和更新的 Vera Rubin 比较,其每 MW 输出 token 的能力依然占优。更特别的是,这些结果使用的还是单 token 预测,没有 speculative decoding,也没有把 prefill…
Z.ai 发布 GLM-5.3-Flash,这是 GLM-5 系列首个原生多模态模型。 它共有 320B 参数,但每次推理只激活 18B。Z.ai 给它的定位很明确:在降低推理成本的同时,把编码、智能体和视觉能力维持在较高水平。官方测试中,它整体超过 GLM-5.2,价格约为后者的十分之一,在部分编码和智能体基准上已经接近 Claude Opus 4.8。 这次变化首先来自模型架构。GLM-5.3-Flash 首次在 GLM 系列中混合使用线性注意力和稀疏注意力:线性注意力处理局部依赖,稀疏注意力负责从长上下文中寻找相关信息。为了继续压低 100 万 token 上下文带来的延迟和显存开销,Z.ai 又加入 IndexPool,对索引向量进行压缩,同时使用 Manifold-Constrained Hyper-Connections(mHC)提高扩展效率。模型预训练使用了约 30T token 的多模态语料。 从规模设计看…
DuckDB 背后的开发团队 DuckLabs 宣布将加入 Amazon Web Services(AWS),相关安排预计在 2026 年 9 月初生效。加入 AWS 后,DuckLabs 团队仍会留在阿姆斯特丹,继续开发 DuckDB、DuckLake、Quack 以及整个 Duck Stack。DuckDB 等开源项目不会因此改变许可证,仍将以 MIT License 免费开源,并继续由非营利组织 DuckDB Foundation 管理。 DuckLabs 成立于 5 年多前,当时团队希望给 DuckDB 提供一个稳定的长期开发环境。公司没有接受风险投资,而是保持创始人与开发团队持有,通过商业合作和技术支持维持运营。此后团队扩展到 30 多人,DuckDB 的使用量也快速增长,目前每天下载量已经超过 100 万次。 随着项目规模扩大,这套模式逐渐遇到压力。DuckLabs 担心,一家规模不大的公司可能无法长期承担越来越…
Qwen3.8-Flash-Next 发布:Qwen4 的新架构提前亮相 Qwen 团队发布了 Qwen3.8-Flash-Next,并开放模型权重。这是一款多模态 MoE 模型,也被视为 Qwen4 架构的一次提前预览。此前 Qwen3-Next 引入的混合架构后来延续到了 Qwen3.5 至 Qwen3.8,这次团队再次选择先把新架构交给社区验证。 Qwen3.8-Flash-Next 的主模型包含 125B 参数,另有 51B N-gram Embedding,每个 token 只激活约 6B 参数。与 Qwen3.7-Plus 相比,其训练成本约为后者的 1/9,同时在代码和办公类任务上取得了更好的成绩。模型原生支持 262,144 tokens 上下文,并可借助 YaRN 扩展至 100 万 tokens。 这次架构调整集中在四个部分。Attention 继续采用 Gated DeltaNet 与全局 Atten…
OpenWorker:让 AI 从聊天窗口走进真实工作流 OpenWorker 是一款开源桌面 AI 协作者。它希望用户直接提出想完成的结果,例如准备客户续约材料、整理会议冲突、分析产品反馈、排查线上事故,甚至检查代码和云环境。接到任务后,OpenWorker 会自行拆分步骤,读取获准访问的文件和服务,再把结果交付为文档、Slack 回复、代码修改建议等可直接使用的内容。 它可以连接 Slack、Gmail、Outlook、Google Calendar、Notion、HubSpot、GitHub、Google Drive、Jira、Linear 等常见工具,也不绑定固定模型。用户可以使用自己的 OpenAI、Anthropic、Google 等 API Key,选择 Kimi、GLM、DeepSeek 等开放权重模型,或者通过 Ollama 在本机运行模型。不同任务之间还可以切换模型。 官网目前把 Security Co…
AI 编程工具正在提高开发效率,但对刚进入行业的人来说,它也可能绕过形成专业能力所需要的训练过程。如今从 AI 编程中获益最多的,往往是已经工作多年、甚至几十年的开发者。他们有足够的经验判断模型输出是否合理,而新人却常被要求一开始就依赖这些工具,同时又被要求具备审查代码、设计架构和识别错误的能力。 这种矛盾在学习阶段尤其明显。Faye 引用的一项针对编程初学者的研究发现,重度依赖生成式 AI 的参与者经常跳过规划和推理,虽然觉得自己学得更快,却容易形成一种“能力幻觉”。反而是那些限制 AI 使用、能够忽略错误建议的人,更容易真正掌握解决问题的方法。经验越少,越难发现模型哪里出了问题;而不知道自己缺什么知识时,甚至很难提出正确的问题。 Faye 把这种被省略掉的困难称为学习中必要的“摩擦”。调试一个没有日志的错误、发现某种实现方式无法扩展、推翻已经写了很久的方案,这些经历看起来低效,却会逐渐形成开发者的直觉和判断力。宾夕法尼亚…
Apple 发布 M6 与 M5 Ultra 两款芯片,并分别将它们用于新款 Mac mini 和 Mac Studio。两款芯片都明显加强了 AI 计算能力,但面向的使用场景不同:M6 更强调日常性能、能效和本地 AI,M5 Ultra 则针对专业创作、科学计算和大型 AI 模型。 M6 是 Apple 首款采用 2 纳米制程的芯片。它配备全新的 12 核 CPU,包括 2 个超性能核心、4 个性能核心和 6 个能效核心。与 M5 相比,多线程性能最高提升 20%;相比 M1,最高达到 2.4 倍。12 核 GPU 的每个核心都集成 Neural Accelerator,AI 峰值计算能力较 M5 提升近 30%,相比 M1 超过 8 倍。新的双 16 核 Neural Engine 峰值算力最高可达前代的 2 倍。 内存方面,M6 最高支持 32GB 统一内存,带宽达到 170GB/s,比 M5 提升 10%。Apple…
Showing the 12 most recent of 20 posts we hold for @synctoai. View and reaction counts are the latest single reading for each post, not a live figure, and a recent post is still accumulating both. A view count marked ≈ was rounded by Telegram before we ever saw it — t.me prints views in full below 1,000 and to three significant figures above, so ≈1,200,000 means somewhere between 1,150,000 and 1,249,999. Unmarked counts are exact. Text is reproduced from the public post preview and truncated for length.
Appears in Telegram’s recommendations for other channels
The reverse of the list above, and a different kind of signal. This does not require this channel to have ever been asked about directly — each row below is a channel we DID ask Telegram about, whose Telegram-generated list happened to include this one. A channel can appear here with an empty list above it, because being named by someone else’s query is independent of having been queried itself.
@OutsightChina · 41,254
Telegram ranks this channel #33 of 58 here — alongside 57 others — read 29 August 2026
This channel appears in 1 seed channel's Telegram-generated recommendation list in total. Each is Telegram’s list for THAT channel, not this one — see how this is measured.
Cite this entry
A live page changes as we take new readings, so a citation should name the measurement it is based on, not just the URL. The line below cites the subscriber count as measured 6 September 2026 — this entry's latest reading, not the date you are reading this.
“Levix 空间站” (@synctoai), 1,105 subscribers as measured 6 September 2026. Telegram Register, tgregister.com/channel/synctoai.
Full measurement history, CC BY 4.0. Every reading this register holds for this entry, not just the latest one, as a dated, downloadable record: CSV · JSON. Free to use with attribution to tgregister.com. Each file carries its own generation timestamp, which is the figure to cite for exactly when the data was retrieved.