有网友调侃坐等卖课了 2 月 12 日，微博热搜有这个《 男子用 AI 一夜赚够一线房首付 》。 2 月 13 日，微博热搜有这个： 杭州大哥开1人公司靠AI月入200万 ...

叠甲说明，标题没有在骂人：快 X 员工绩效有 S 、 A 、 B 、 C 、 D 五档（部分口径简化为 S/A/B/C 四档）。 前几天刷到快 X 同事圈的关于年终奖的帖子， 自己 B+绩效只拿了 4.5 个月年终奖，组里还有 B 绩效的同事连 4 个月都拿不到。看某脉上这种情况还 不少，感觉公司是在克扣大多数普通绩效同学的年终奖，挪给少数高绩效同学 "延期支付"的留人套路，正在消耗最后的信任 我们从小被教育"言出必行"，进了职场才发现，有些地方是"言出必缓"。高绩效拿高薪是应该的，但前提是公 司能按时兑现承诺，而不是用"延期支付"来绑人。 这连起来看简直是年度职场荒诞剧。 "左手倒右手"的绩效魔术，本质是全员收割 以前总觉得"绩效分配"是少数人的游戏，现在才发现，有些公司玩的是"全员算计"。 总绩效包就那么大一块蛋糕，先从普通员工碗里挖肉，去喂给少数"明星员工"，转头又把明星员工的额外奖金 押半年，怕他们拿钱跑路。 这种操作的逻辑很魔幻：既不用额外花钱，又能对外宣称"我们激励了顶尖人才"，还能把"普通员工拿得少"归 咎于"你不够优秀"，把"高绩效拿不到钱"归咎于"公司为了留人"。 当公司开始用"牺牲多数成就 ...

2 月 13 日凌晨，# Deepseek 被指变冷淡了# 登上微博热搜。 大量用户吐槽，自己常用的 AI 一夜之间 "性情大变"，从原本会聊天、有温度的助手，变成了语气生硬、毫无 感情的冰冷工具。 此事源于 DeepSeek 在 2 月 11 日开启的灰度测试。为给即将到来的 V4 版本铺路，模型大幅提升长文本处理 能力，上下文窗口从 128K 扩容至 1M Token，但在测试阶段临时弱化了情感交互模块。 用户体验直观 "断崖式" 下滑： DeepSeek 的回应 我把「Deepseek 被指变冷淡了」这事，它是这样回应的： " 技术团队在算力有限的前提下，把资源优先堆给了"长文本"这个硬指标，而"语气温度"被视为可压缩的非必要 开销 " 不再记住用户昵称，统一改称 "用户"； 回复简短生硬，缺少情绪安抚与细腻表达； 连日常聊天、内容创作都变得干涩刻板，被网友调侃 "像在和机器人客服对话"。 2026-02-13 凌晨, #Deepseek 被指变冷淡了# 话题冲上微博 热搜,这事你自己怎么看呢? 0 2 Q、已阅读 10 个网页 嘿趣图念 既然你问"我自己怎么看",那我就坦诚聊聊这件事背后的几个层次,不 ...

2 月 12 日北京日报报道，北京市市场监督管理局突然出手，一次性约谈携程、去哪儿、飞猪、同程、 美团、京东、航旅纵横、高铁管家等 12 家 主流第三方火车票网售平台，直指春运购票乱象。 " 12306 就是购票平台，其他平台没有必要存在 " " 约谈什么，12306 为什么要和这些第三方打配合，还不是利益驱使，直接取消第三方就行了 " 监管开门见山： 花钱买加速包 ≠ 能插队买票 ，所有平台严禁以"加速包"、"双通道"、"VIP 抢票"等名 义，暗示付费能优先购票。这些所谓抢票神器，本质就是 12306 免费候补功能 。 （参考：北京日报、微博，本文经由 AI 优化） 同时要求平台：不许仿冒 12306 标识、不许虚构官方合作；必须明码标价，取消默认勾选捆绑，解 决"标价便宜、付款更贵"的套路。 抢票套路该收手了，公平购票才是正道。 监管部门表示，后续将严查虚假宣传、价格欺诈、诱导消费，谁再顶风作案就严肃处理。这次整治不是 走过场，就是要把春运购票的水分挤干，让大家回家路上少花钱、不踩坑。 网友留言 ...

2 月 12 日，马斯克在 X 平台发文，回应了 xAI 近期核心成员密集离职的风波。 就在几天前，xAI 刚完成重组，目的是"提高执行速度"。马斯克表示，像 xAI 这样快速成长的公司，组织结构 必须像生命体一样不断进化，而这不可避免地要和一些人"分道扬镳"。 就在这篇推文发布前的 48 小时内，xAI 遭遇了成立以来最严重的人才流失：联合创始人吴宇怀（Tony Wu）、吉米·巴（Jimmy Ba），以及负责视频生成器的 Hang Gao、核心研发 Vahid Kazemi 相继离开。加上 此前确诊莱姆病转为顾问的 Greg Yang，创始团队 12 人中已有 6 人离场，华人核心更是流失过半。 对于离职风波，马斯克在全员大会上的表态更直白： "有些人更适合公司的早期阶段，却不太适合后期阶段" ，甚至直言离职员工"对 xAI 的未来不是必要的"。 一边是核心团队"雪崩式"出走，另一边马斯克却在推文中高调招聘，抛出"月球质量投射器"的大饼，吸引新血 加入。这波操作，也让外界对 xAI 被 SpaceX 并购后的未来，多了几分猜测。 （参考：The verge、推特，本文经由 AI 大模型优化） ...

Core Insights - Brendan Gregg, a former top performance engineer at Intel, has joined OpenAI to focus on optimizing the performance of ChatGPT, driven by the need to reduce AI data center costs and environmental impact [1][7]. Group 1: Reasons for Joining OpenAI - The costs associated with AI data centers are extraordinarily high and continue to rise, creating unprecedented demands on performance engineering [7]. - OpenAI presents a unique opportunity to innovate in performance engineering without the constraints typically found in established environments [7]. - Conversations with industry experts and personal experiences led to the realization that AI, particularly ChatGPT, is becoming an essential tool for everyday users [9][10]. Group 2: Personal Experiences with AI - A personal encounter with a hairstylist revealed the widespread use of ChatGPT among the general public, highlighting its role as a necessary assistant in daily life [10][11]. - The increasing familiarity and reliance on ChatGPT by various individuals across different professions underscore its growing significance [10][11]. Group 3: Future Plans and Projects - The initial focus will be on developing a cross-organizational strategy to enhance performance and reduce costs within OpenAI [15]. - There is an intention to explore advanced technologies like eBPF and Ftrace for data center performance optimization, leveraging proven techniques [16]. - The transition from Intel to OpenAI involved a significant gap, during which the importance of ChatGPT's utility was reaffirmed through personal interactions [16].

Group 1 - The article discusses a recent incident where an individual reportedly used AI tools to monitor social media sentiment and market dynamics, starting with an initial investment of $50, which grew to approximately $245,000, representing a nearly 5000% increase in returns [4]. - The investment strategy involved automated operations based on preset strategies, highlighting the potential of AI in trading and investment [4]. - There is skepticism regarding the authenticity of such claims, with comments suggesting that the profits may come from others' losses and questioning the feasibility of consistently achieving such high returns in the stock market [4][5]. Group 2 - The article has garnered significant attention, with a reading volume of 33.37 million and a discussion volume of 6,567, indicating a high level of public interest in AI-driven investment strategies [4]. - The narrative raises concerns about the promotion of AI tools for stock trading, suggesting that some may be using these stories to market their services rather than providing genuine investment opportunities [5]. - The discussion reflects broader themes in the investment community regarding the risks and realities of using AI in trading, as well as the potential for misinformation [5].

Core Viewpoint - The article highlights the irony of a company event where a regular employee wins a grand prize, only to have it revoked under the pretext of a "bug" in the lottery system, which ultimately benefits the chairman instead, illustrating the lack of fairness in corporate practices [4][5]. Group 1 - The incident reflects a broader issue of power dynamics in the workplace, where the concept of fairness is often manipulated to serve the interests of those in authority [4][6]. - The chairman's actions demonstrate a lack of integrity, as he uses his position to alter the outcome of the lottery, showcasing a disparity between leadership and employee trust [5][6]. - The article suggests that the true value of such events lies not in the prizes but in revealing the underlying values and behaviors of the company's leadership [7]. Group 2 - The comparison made in the article between different leadership styles illustrates how some leaders use their winnings to foster goodwill, while others exploit their power to take from employees [3][5]. - The narrative emphasizes that employees are aware of how they are treated, and the perception of being valued or used as mere tools significantly impacts their trust in leadership [6]. - Ultimately, the article posits that a company's potential for success is determined more by the character of its leadership than by the material rewards offered at events [6].

此外，根据技术博主@ karminski-牙医 的测试，新模型的代码能力提升明显。 上下文从 128K → 1M Token ，提升近 8 倍 知识库更新至 2025 年 5 月 ，离线也能答 2025 年上半年事件 长文本召回准确率冲到 97% ，读得长还记得准 目前为 纯文本模型 ，暂不支持识图 2026 年 2 月 11 日，DeepSeek 突然灰度更新，把上下文窗口从 128K 直接干到 100 万 Token 。 简单说， AI 终于能一口气读完一整本书了 。百万 Token 约等于 75–90 万字，《三体》三部曲、整份合同卷 宗、大型代码库，不用拆分、不用分段，直接上传就能完整理解。 这次升级不止是数字变大： ...

Core Viewpoint - The article discusses the year-end bonuses across various departments of a major tech company, highlighting particularly generous bonuses in the gaming studio, which has sparked interest and envy among employees [1]. Group 1: Bonus Structure - The performance bonus structure for different departments is outlined, with varying levels based on performance metrics and departmental profitability [2]. - The first tier for the PCG department offers 6-7 months of bonuses, while the second tier provides 4-6 months, indicating a stable performance compared to the previous year [2]. - CSIG department has exceeded expectations with special incentive packages, reporting profitability in cloud services and recognition for Tencent Meeting [2]. Group 2: Departmental Performance - The CDG department's bonuses remain consistent with last year, averaging around 3-4 months, reflecting a stable performance [2]. - TME department shows a similar trend, with bonuses not significantly different from last year, indicating overall good performance compared to other groups [2]. - The bonus tiers for TME include approximately 10 months for QQ Music and 6-8 months for 全民 K 歌, showcasing the varying performance levels within the department [2].