🔥 Search Hot Tweets

Search and analyze hot tweets from KOL accounts within 6 hours. Use SoPilot plugin to quickly comment and occupy the comment section.

Real-time Hot Tweet Analysis

dontbesilent
69.3Kfo
dontbesilent@dontbesilent12· 21h ago发布

闪电说更新完之后 nb 了 长按是语音转文字(现在用的豆包模型 短按可以直接和 AI 对话(简单回复现在用 qwen plus 足够了 https://t.co/jkETYq91WQ

53
4
14
9.3K
Data updated 8h ago
Viral Probability
61%
Predicted Views
13.0K
Est. 100 views for your reply
Ding
144.2Kfo
Ding@dingyi· 21h ago发布

claw 生态太恐怖了,每天都涌现出好多新产品新玩法。一下子激活了整个行业。最搞笑的是,在小龙虾诞生前不久,Meta 花巨资收购了那个。。。

258
15
36
89.9K
Data updated 9h ago
Viral Probability
100%
Predicted Views
344.0K
Est. 700 views for your reply
underwood
13.9Kfo
underwood@underwoodxie96· 21h ago发布

{ "subject": { "description": "An adult Korean woman with refined, clean facial lines posing casually by a wooden sliding glass door, wearing a navy blue lace lingerie set, calm half-lidded gaze with a quiet, intimate confidence.", "mirror_rules": [ "Not a mirror selfie; normal camera capture", "No mirrored text artifacts", "Keep facial proportions realistic and consistent" ], "age": "adult (20s)", "expression": { "eyes": { "look": "soft half-lidded gaze", "energy": "calm, subtly teasing, controlled", "direction": "toward the camera" }, "mouth": { "position": "lips gently closed", "energy": "relaxed, faint pout" }, "overall": "quiet intimacy, warm composure" }, "face": { "preserve_original": false, "makeup": "K-beauty natural glam: sheer dewy base, straight soft brows, subtle inner eyeliner, light aegyo-sal highlight, muted gradient rosy lips" } }, "hair": { "color": "dark brown-black", "style": "see-through bangs, the rest tied back in a low ponytail or low bun", "effect": "soft wispy strands framing the cheeks, slightly glossy" }, "body": { "frame": "slim with soft curves", "waist": "narrow waist", "chest": "full bust, natural cleavage", "legs": "toned thighs, long legs", "skin": { "visible_areas": "arms, shoulders, torso, hips, legs", "tone": "fair warm-neutral", "texture": "smooth and supple, satin-like to the touch, faint natural pores", "lighting_effect": "gentle rim light on edges, soft highlight bloom, slightly dewy sheen" } }, "pose": { "position": "standing, leaning lightly against the wooden door frame", "base": "hips subtly angled, weight on one leg", "overall": "both arms raised above the head, elbows bent, creating an arched silhouette" }, "clothing": { "top": { "type": "lace bra", "color": "navy blue", "details": "floral lace texture, small bow accents near the straps, structured cups", "effect": "elegant, delicate lace catching the backlight" }, "bottom": { "type": "lace panties with thin side straps", "color": "navy blue", "details": "matching lace front, strappy side ties/loops, minimal coverage", "effect": "fine lace pattern visible, crisp edges against skin" } }, "accessories": { "prop": "a blue-and-white patterned yukata/robe draped behind her shoulder, large floral motifs, slightly rumpled fabric" }, "photography": { "camera_style": "candid natural-light portrait with a soft compact-camera feel", "angle": "slight dutch tilt, near eye-level but subtly low, framed by the sliding door", "shot_type": "three-quarter body portrait, subject centered-right in frame", "aspect_ratio": "1:1 (almost square)", "texture": "mild film-like grain, gentle haze, slight highlight bloom, not overly sharp", "lighting": "strong outdoor backlight through glass, diffused and warm, soft rim light, slightly overexposed background", "depth_of_field": "moderate; subject clear, background softened by glass haze" }, "background": { "setting": "traditional wooden room with a sliding glass door opening to a small garden", "wall_color": "natural wood tones", "elements": [ "wooden sliding door frame", "glass panel with faint haze/condensation", "garden rocks and greenery outside", "wooden fence in bright sunlight", "bamboo blind texture along the right edge", "a woven basket partially visible in the lower-left area" ], "atmosphere": "quiet afternoon, warm sun spilling in, private and serene", "lighting": "sunlit exterior is bright and blown-out, interior is softly shaded with gentle gradients" }, "the_vibe": { "energy": "soft, intimate, quietly bold", "mood": "sun-warmed stillness, a private pause between moments", "aesthetic": "realistic social-photo sensuality with subtle K-beauty polish, imperfect and human", "authenticity": "slight camera tilt, natural skin texture, tiny exposure mistakes, lived-in fabric wrinkles", "intimacy": "like she just slipped the robe off after returning from a warm bath, the window still faintly fogged", "story": "a calm room, garden light, and a confident breath held for the shutter click", "caption_energy": "low-voice, close-distance, lingering sunlight" }, "constraints": { "must_keep": [ "adult Korean woman", "navy blue lace lingerie set with bow accents", "arms raised above head, leaning against wooden door frame", "blue-and-white yukata/robe draped behind shoulder", "strong backlight with hazy glow and slight overexposure outside", "traditional wooden sliding glass door and garden rocks/greenery" ], "avoid": [ "explicit nudity", "pornographic framing", "overly plastic skin", "extreme body exaggeration", "heavy studio lighting", "text/logo overlays" ] }, "negative_prompt": [ "child", "teen", "loli", "underage", "explicit nude", "porn scene", "spread legs", "hyper exaggerated breasts", "tiny waist distortion", "plastic skin", "over-smoothing", "harsh flash", "studio backdrop", "perfect symmetry", "extra fingers", "bad hands", "deformed anatomy", "watermark", "logo", "readable text" ] } More real-life image prompts: https://t.co/w2tp3YGqUy

128
9
8
4.9K
Data updated 9h ago
Viral Probability
61%
Predicted Views
10.0K
Est. 100 views for your reply
铁锤人
80.5Kfo
铁锤人@lxfater· 21h ago发布

为什么总有那种几千粉丝的博主教别人 起号 写文章 分析别人策略,比如我 我本来不想多说的,但是今天给你们分享一下 一句话,重复成功过三次的人才算懂 很明显我不是,所以分析我显得很新手 先会学,再说教😂 重复成功三次不是我说,达利欧说的

42
0
37
6.8K
Data updated 9h ago
Viral Probability
58%
Predicted Views
9.0K
Est. 100 views for your reply
向阳乔木
91.8Kfo
向阳乔木@vista8· 22h ago发布

这个Github 库是 X 上哪个朋友写的,太酷了! 自带Gemini 和 Codex的 MCP。 让Claude Code组织分工,一次性完成复杂编程,断了也能恢复(有Spec文件)。 比如让Gemini写前端,Codex修bug,充分利用模型优势。 哪怕不编程,ChatGPT订阅送的Codex和谷歌的Gemini Cli都要被榨干啊。 地址见评论

223
33
4
20.5K
Data updated 9h ago
Viral Probability
74%
Predicted Views
32.0K
Est. 1.1K views for your reply
𝙩𝙮≃𝙛{𝕩}^A𝕀²·ℙarad𝕚g𝕞
16.1Kfo

delegation stack theory recursive self-improvement loops 最近的论文明显到了一个新阶段,这些关键词一股脑从我脑海里冒出来 https://t.co/dCgM5o52D5

63
9
1
3.8K
Data updated 9h ago
Viral Probability
58%
Predicted Views
8.0K
Est. 400 views for your reply
0x卡卡撸特
8.8Kfo
0x卡卡撸特@0xkakarot888· 22h ago发布

卡卡的 X 增长陪跑营 tg 群限时开放,要来的速进,48小时关闭入口。 https://t.co/X6ne9IVfBF https://t.co/HM6ve4IYmU

18
1
13
4.6K
Data updated 10h ago
Viral Probability
47%
Predicted Views
10.0K
Est. 100 views for your reply
Leo Xiang
17.3Kfo
Leo Xiang@leeoxiang· 22h ago发布

一杯咖啡之旅。 by Claude code + Remotion + Excalidraw Excalidraw + Remoiton 用来教育以及数据可视化类视频效果太好了。 https://t.co/WIU3kBNrZM

120
19
7
10.8K
Data updated 10h ago
Viral Probability
67%
Predicted Views
15.0K
Est. 100 views for your reply
Xiangyu 香鱼🐬
22.4Kfo
Xiangyu 香鱼🐬@XianyuLi· 22h ago发布

哈哈,好有道理 https://t.co/dInTsGyOIp

20
2
6
2.2K
Data updated 10h ago
Viral Probability
54%
Predicted Views
5.0K
Est. 100 views for your reply
dontbesilent
69.2Kfo
dontbesilent@dontbesilent12· 22h ago发布

一开始让 claude code 设计了 Agent Team,试图帮我通过一小段文案碎片,把这段内容直接变成一个可发布的选题 结果整理出来的东西太长,5 个 Agent 总共写了 4700 字,我这边阅读压力太大了,迭代变慢 让 claude code 用德鲁克管理学优化了 Agents 之后,现在 Agent Team 的产出的第一部分是:一页纸执行摘要,把我需要看的信息压缩到了 700 字左右 德鲁克真牛逼😂

64
5
2
9.3K
Data updated 10h ago
Viral Probability
57%
Predicted Views
13.0K
Est. 400 views for your reply
dontbesilent
69.2Kfo
dontbesilent@dontbesilent12· 23h ago发布

设计 Agent Team 的时候,感觉自己不太会安排多个角色协作 想到德鲁克应该写过这些内容,看看德鲁克的书应该就能解决 可是他的内容里面又考虑了太多人性因素,不是专门为 AI Agent 打造的,还得为 Agent 改造一下 然后突然反应过来,其实一句提示词就能解决: 设计 Agent 协作系统时,应用德鲁克管理学原则:目标管理、分工专业化、决策权下放、有效沟通、知识工作者管理,忽略人性相关部分(激励、情感) —— 顺便,从这个提示词里面也发现,过度关注激励和情感,就是创业陷阱

31
1
6
9.3K
Data updated 11h ago
Viral Probability
55%
Predicted Views
13.0K
Est. 100 views for your reply
Li | Design for Startups
9.4Kfo
Li | Design for Startups@lizengco· 23h ago发布

I lost a client who genuinely loved my work, and it was the best business lesson I ever had. Today, I’ll share the painful pricing lesson that fixed my boundaries forever. When I first started my design studio, I wanted things to feel simple. I hated writing proposals, and I wanted clients to say “yes” fast. So I made everything a flat fee. Website? One price. Landing page? One price. Easy, clean, no back and forth. But I made one rookie mistake. I priced the “thing,” not the variables inside the thing. Because a website is never just a website. It’s page count, messaging, visuals, layout, interaction, animation, and revisions. And those “small adds” pile up fast. This client came from a friend's introduction. He loved our style and was happy with the quality. When he heard “flat fee,” he locked onto it like a promise. In his mind, a flat fee meant unlimited. And that’s where the slow bleed started. It began harmless. “Can we add a few more pages?” “Can we add more images?” “Can we tweak this section again?” Each request sounded small. But each request cost real time. And I kept saying yes. Then I tried to draw a line: this is beyond scope, we’ll need to add hours or add a fee.” The client pushed back immediately: “But you said it was a flat fee, why are you charging more now?” To be honest, I compromised too many times. It felt like I was the only one bending. So I got frustrated, and the collaboration got tense. In the end, he still loved the final website. But he never came back for more work. Losing that long-term client hurt. Not because my work wasn’t good. Because my pricing system wasn’t fair. That was the moment I realized two things: Pricing is not math. Pricing is boundaries. Early entrepreneurship is never perfect. Sometimes you lose money. Sometimes you lose clients. Sometimes you lose sleep. But the lesson is the real profit. That one client forced me to become precise. And today, my contracts are crystal clear because of it. What are some early lessons you have learned building a business?

32
0
6
1.5K
Data updated 11h ago
Viral Probability
55%
Predicted Views
4.0K
Est. 100 views for your reply
Bear Liu
112.4Kfo
Bear Liu@bearliu· 23h ago发布

提升品味,2026年要读一读这三本书。 不只是对于设计师,内容创作者和产品人,以及有追求的程序员都适合。 https://t.co/Vgkfke9MH1

79
13
7
9.3K
Data updated 11h ago
Viral Probability
63%
Predicted Views
13.0K
Est. 100 views for your reply
Wey Gu 古思为
15.1Kfo
Wey Gu 古思为@wey_gu· 1d ago发布

是基于图上的聚类算法哈,能找到全局的知识聚集性,v0.6 会主动去做,之前的话可以手动触发 参考文档: https://t.co/A4QZqyqJvj https://t.co/ozjaoedPaf

38
5
0
3.8K
Data updated 11h ago
Viral Probability
51%
Predicted Views
8.0K
Est. 400 views for your reply
Kai
28.4Kfo
Kai@real_kai42· 1d ago发布

之前大家开玩笑 “anthropic CEO 在百度实习的时候经历了什么,让他这么狠” 现在我想说 “他到底在百度实习了多久?怎么都学会了?”

416
28
10
101.7K
Data updated 12h ago
Viral Probability
100%
Predicted Views
339.0K
Est. 2.2K views for your reply
徐冲浪
124.2Kfo
徐冲浪@cyrilxuq· 1d ago发布

https://t.co/KnVR8mcv2L

80
8
14
12.9K
Data updated 11h ago
Viral Probability
63%
Predicted Views
17.0K
Est. 100 views for your reply
forecho📈
38.4Kfo
forecho📈@caizhenghai· 1d ago发布

想不通,美团就请我喝过一杯奶茶,既然能亏这个样子 https://t.co/dgZPfjX7JM

137
3
30
50.1K
Data updated 12h ago
Viral Probability
54%
Predicted Views
91.0K
Est. 100 views for your reply
dontbesilent
69.2Kfo
dontbesilent@dontbesilent12· 1d ago发布

用 claude code 的前提是:你的工作流程本来就基于文件系统 用 claude code Agent team 的前提是:你的工作流程本来就需要多个角色协作 用 prompt 产出文字/图片/视频的前提是:你的产出背后有显性的规则可以写出来 这三个前提,都对人的要求很高 而那些被我们称之为“割”的业务,就是展示了一个炫丽的结果,却抛开这些前提不谈,直接卖你一个:claude code 远程安装 差不多就是给没有驾照的人卖车载充电器 而我最近看了一些抄我内容的人(视频号居多),他们自己都没有工作流程,一顿胡搞,然后靠帮人安装工具变现 我才发现这些卖家自己也没有驾照 然后这些人竟然还有群,还有私域直播,你很难想象一群没有驾照的人,可以组织起来车友俱乐部线下聚会

66
4
5
7.1K
Data updated 12h ago
Viral Probability
57%
Predicted Views
10.0K
Est. 100 views for your reply
CuiMao
28.8Kfo
CuiMao@CuiMao· 1d ago发布

我买的Seedance2最高档位,根本不用排队,别人还在苦苦等待无限抽卡的时候,你已经上传好等待收割流量了。这就是 AI 时代的效率。 https://t.co/b5DPzJlxJo

35
1
19
6.5K
Data updated 1d ago
Viral Probability
65%
Predicted Views
29.0K
Est. 100 views for your reply
Lexi 勒西
12.5Kfo
Lexi 勒西@lexi_labs· 1d ago发布

西方典故的三大出处: 1. 圣经 2. 莎士比亚 3. 希腊神话 这三个读过了,能理解 90% 的英文典故。

78
11
5
5.5K
Data updated 1d ago
Viral Probability
74%
Predicted Views
24.0K
Est. 300 views for your reply
Kai
28.4Kfo
Kai@real_kai42· 1d ago发布

emmm,不是说春节前很热闹么,感觉也没有想象中的热闹。。 大的到底来不来了

26
0
16
9.1K
Data updated 1d ago
Viral Probability
53%
Predicted Views
39.0K
Est. 200 views for your reply
underwood
13.9Kfo
underwood@underwoodxie96· 1d ago发布

seedance 2.0 Crazy McDonalds https://t.co/YEWFISygJt

103
13
10
6.0K
Data updated 1d ago
Viral Probability
80%
Predicted Views
29.0K
Est. 200 views for your reply
歸藏(guizang.ai)
124.4Kfo
歸藏(guizang.ai)@op7418· 1d ago发布

年前最后一天了,准备放假。 顺便感谢一下大家的新年礼盒,祝 2026 年中国 AI 越来越好! 谢谢各位:Rico 有三猫、观猹、WaytoAGI、小红书、通义、阿里云、腾讯、百度、字节、Trae、即梦、360、可灵、快手研发线、VAST、爱诗科技、智谱、知乎、电子工业出版社 https://t.co/0ALbBood2P

43
0
18
5.1K
Data updated 1d ago
Viral Probability
63%
Predicted Views
23.0K
Est. 100 views for your reply
Tiezhen WANG
9.9Kfo
Tiezhen WANG@Xianbao_QIAN· 1d ago发布

AIDC-AI/Ovis2.6-30B-A3B is out on @huggingface https://t.co/CnjnbP1gLk https://t.co/21vNJCYd7A

29
4
0
3.6K
Data updated 1d ago
Viral Probability
61%
Predicted Views
11.0K
Est. 700 views for your reply
Asuka小能猫
10.9Kfo
Asuka小能猫@AsukaOdysseus· 1d ago发布

有人真的是身在福中不知福,让这么漂亮的女孩子等他。 说明什么?说明找到了比这更上瘾的事。 我也要对工作上瘾,你能996,我身体好可以006。 https://t.co/KyQauzQjhA

48
0
7
9.1K
Data updated 1d ago
Viral Probability
57%
Predicted Views
20.0K
Est. 100 views for your reply
Guohao Li 🐫
12.1Kfo
Guohao Li 🐫@guohao_li· 1d ago发布

marked exactly one month since this tweet. guess what? we didn’t get killed. we went from almost zero revenue to $500K in revenue last month and we’re profitable as a small team. sometimes a fast, small decision can make a huge impact. what I learned: never be afraid of making bold decisions

106
5
15
10.9K
Data updated 1d ago
Viral Probability
67%
Predicted Views
31.0K
Est. 100 views for your reply
Orange AI
159.4Kfo
Orange AI@oran_ge· 1d ago发布

https://t.co/PNhCJwlKVd

72
10
17
30.8K
Data updated 1d ago
Viral Probability
100%
Predicted Views
234.0K
Est. 1.1K views for your reply
Tiezhen WANG
9.9Kfo
Tiezhen WANG@Xianbao_QIAN· 1d ago发布

The AI researcher has woken up https://t.co/QUk2Rq2DcW https://t.co/7MssoomKeM

28
1
1
2.7K
Data updated 1d ago
Viral Probability
53%
Predicted Views
6.0K
Est. 300 views for your reply
向阳乔木
91.3Kfo
向阳乔木@vista8· 1d ago发布

Slack前CEO说过一句话特别到位:"创新的唯一真实衡量标准,是人的行为是否改变了。" 左:把AI工具塞进旧业务模式 右:以AI为基础重新设计业务 "如果用AI重新设计一个XX公司,AI做它擅长的,人做人擅长的,分别是什么?"

26
2
3
3.1K
Data updated 1d ago
Viral Probability
59%
Predicted Views
7.0K
Est. 400 views for your reply
dontbesilent
68.7Kfo
dontbesilent@dontbesilent12· 1d ago发布

很多人误以为上学和读书是投资行为 但在多数情况下,这俩都更接近于消费行为 除非你有一个明确的商业模式,可以把你的阅读量、学历商品化 否则你花钱上学读书,和你花钱喝杯奶茶没有本质区别 大家觉得上学是投资的原因可能有二: 1、上学痛苦难受 2、日后能赚更多钱 可是我花钱点了个外卖,如果做的太咸了,我也难受,不能因为我难受,点外卖就成了投资? 「日后赚钱」这个事就更扯了,我们讲基础逻辑:先有目的,后有手段 你连你卖啥都不知道,你凭啥知道读这本书就可以帮助到事业? 就算真帮到了,也是瞎猫碰上死耗子,我喝奶茶就不能发现商机吗 (つ✧ω✧)つ🧋

124
9
24
11.8K
Data updated 1d ago
Viral Probability
73%
Predicted Views
45.0K
Est. 100 views for your reply
Bear Liu
112.4Kfo
Bear Liu@bearliu· 1d ago发布

我靠,效率太高了!用 Claude Code 将我拖了两个月的项目网站,今天一下午就做完并上线。中间还去海里游了一个泳,做了个饭,带了组娃。 唯一阻挡进度的就是 Token需要重置,每次用完需要两三个小时的恢复期。 https://t.co/RdMgSTpJk6

28
0
10
5.6K
Data updated 1d ago
Viral Probability
58%
Predicted Views
13.0K
Est. 100 views for your reply
Asuka小能猫
10.9Kfo
Asuka小能猫@AsukaOdysseus· 1d ago发布

英文推特被越来越多的大佬关注后,只敢在中文小号shitpost了。。但不知不觉中文小号也越来越大了。

27
0
5
2.6K
Data updated 1d ago
Viral Probability
46%
Predicted Views
6.0K
Est. 100 views for your reply
Tony出海
21.1Kfo
Tony出海@iamtonyzhu· 1d ago发布

原来大家用的VPN也是经过18道中转的, 以前搞不明白明明节点选的是美国,实际测ip却在欧洲,是服务商层层套壳中转的结果。所以大家都找住宅IP。 Claude Max的API服务行业也是18道中转,很多都掺杂各种薅羊毛来的逆行API,几乎很少能用上官方纯血max,降智和极不稳定。 我帮过不少人代买Claude Max,但是A社封号堪比GFW,成本太高了,一中断工作,太耗费时间心力了。 现在我都是推荐直接用 https://t.co/Hr3qcP3lLD 企业级API服务商,稳定正规靠谱。 Seedance 2.0让老外垂涎欲滴, 国产模型爆发,模型这么多,该用哪家呢? 全都要,ZenMux聚合平台,一站式使用国外最强模型和国产最新模型。那一堆国产模型太分散了,账号这里一站式体验到。

68
9
12
17.9K
Data updated 1d ago
Viral Probability
61%
Predicted Views
63.0K
Est. 300 views for your reply
Li | Design for Startups
9.4Kfo
Li | Design for Startups@lizengco· 1d ago发布

Looking into more details. I am still a fan of beige background with dark green :) https://t.co/o4OovEFslc

34
0
2
1.1K
Data updated 1d ago
Viral Probability
54%
Predicted Views
3.0K
Est. 200 views for your reply
Kai
28.4Kfo
Kai@real_kai42· 1d ago发布

我让车上的小爱同学给我倒计时,小爱直接给我整了段 js 代码 https://t.co/pwOf7xgI0Q

400
7
28
52.0K
Data updated 1d ago
Viral Probability
100%
Predicted Views
333.0K
Est. 1.0K views for your reply
Ding
144.1Kfo
Ding@dingyi· 1d ago发布

哈哈哈连 Musk 都看不下去了。Anthropic 的 CEO 太尼玛恶心了,至于那几个华裔员工,也是帮凶。

319
14
30
137.0K
Data updated 1d ago
Viral Probability
100%
Predicted Views
1.5M
Est. 4.5K views for your reply
Frank Wang 玉伯
26.6Kfo
Frank Wang 玉伯@lifesinger· 1d ago发布

不少产品经理看了这条不舒服 何必呢 人就是人 有 AI 加持后 懂 Claude Code 的产品经理 就不是产品经理了 身份认识转变后 前途无量 如果依旧对产品经理念念不忘 那确实 基本完了

25
3
21
10.7K
Data updated 1d ago
Viral Probability
60%
Predicted Views
14.0K
Est. 100 views for your reply
吕立青_JimmyLv 2𐃏26
17.6Kfo
吕立青_JimmyLv 2𐃏26@Jimmy_JingLv· 1d ago发布

确实这次去香港开卡,只有中银香港没成功,😂 线上App+出入境记录PDF+港澳通行证/内地身份证+酒店WiFi+扫脸就搞定了,全程1h不到就能开五张卡,大家感兴趣我可以出详细教程~ https://t.co/0xkYpCgFFN

190
27
40
42.2K
Data updated 1d ago
Viral Probability
70%
Predicted Views
82.0K
Est. 100 views for your reply
马东锡 NLP
36.4Kfo
马东锡 NLP@dongxi_nlp· 1d ago发布

我用 MiniMax-M2.5 升级了我的 AI 论文阅读伙伴。 这个应用自动抓取并分析了近一年 1 万篇 AI 论文,建立起数据基础,用来理解 AI 领域的最新的主题和方向,并结合时间序列计算热度趋势。 在这个基础上,每天产生的每一篇新论文,它的主题归属、潜在价值与趋势变化,就非常清晰地展示在眼前。 M2.5 端到端的完成度超强,多个复杂计算和任务完成的非常出色,并且能把复杂的计算做成非常顺手的交互。 Kudos to @MiniMax_AI ! 对 MiniMax 的好感,最早来老罗和闫俊杰那期播客,感叹头部模型公司的老板也可以这么真诚。 我只用三家 AI 产品,MiniMax 是唯一的国产 AI 伙伴。

167
21
20
21.0K
Data updated 1d ago
Viral Probability
70%
Predicted Views
30.0K
Est. 100 views for your reply
Orange AI
159.4Kfo
Orange AI@oran_ge· 1d ago发布

今天这篇文章突破100万曝光了 这篇文章引起了很多有价值的讨论 这些讨论也让我更严肃地思考互联网世界和 Agent 世界 相比这篇中性而相对冷静的文章,英文世界更激进,更焦虑,甚至开始用疫情来类比现状,开始出现很多吹哨人 我也不知道怎么评价了 昨天结束了全面的工作,晚上睡得很香

27
1
3
22.9K
Data updated 1d ago
Viral Probability
54%
Predicted Views
32.0K
Est. 900 views for your reply
宝玉
181.2Kfo
宝玉@dotey· 1d ago发布

Spotify 语出惊人:得益于 AI,我们最顶尖的开发者从去年 12 月起就没亲手写过一行代码 From:TechCrunch AI 编程是否已经到达了引爆点?对于 Spotify 来说,答案似乎是肯定的。 在本周举行的第四季度财报电话会议上,Spotify 分享了一个令人咋舌的消息:公司里最优秀的开发者“自去年 12 月以来,就没有写过哪怕一行代码”。这番言论出自 Spotify 联合首席执行官 Gustav Söderström 之口,背景正是该公司正在利用 AI 全面加速产品研发。 值得注意的是,Spotify 指出他们在 2025 年全年为这款流媒体应用发布了超过 50 项新功能和改进。而且就在最近几周,他们又密集推出了一系列新功能,比如由 AI 驱动的“提示词播放列表”(Prompted Playlists)、针对有声读物的“页面匹配”(Page Match),以及“关于这首歌”(About This Song)。 为了实现如此惊人的产品迭代速度,Spotify 告诉分析师,他们的工程师正在使用一套名为“Honk”的内部系统。这个系统允许利用生成式 AI——具体来说是 Claude Code——来实现远程、实时的代码部署。 Söderström 举了一个非常具体的例子,画面感极强: “想象一下,一位 Spotify 的工程师在早上通勤的路上,只需要掏出手机,在 Slack (一款类似飞书、钉钉的企业办公协作软件) 上告诉 Claude 去修复一个 Bug 或者给 iOS 应用加个新功能。当 Claude 完成工作后,工程师的手机 Slack 上就会收到一个新版本的应用安装包。他可以直接测试,然后合并发布到生产环境——而这一切,甚至发生在他还没踏进办公室大门之前。” Spotify 认为,这套系统在加速编程和部署流程方面功不可没,效率提升是“巨大”的。 “我们预见,这绝不是 AI 开发模式的终点,而仅仅是个开始,”Söderström 说道。 这位高管还吹捧了 Spotify 构建独特数据集的能力,这是其他大语言模型(LLM)无法像抓取维基百科那样轻易将其变成通用商品的。他解释说,因为音乐相关的问题并不总有一个绝对的事实答案。 举个例子,如果你问“什么是健身音乐?”,不同的人会给你完全不同的答案,这往往取决于他们住在哪里。美国人总体上偏爱嘻哈音乐,但也有数以百万计的人喜欢伴着死亡金属(death metal)锻炼;很多欧洲人喜欢电子舞曲(EDM),而许多斯堪的纳维亚人 (北欧人) 则对重金属情有独钟。 “这是我们目前正在构建的一个数据集,其他人都没有真正在做。这种规模的数据集是独一无二的。而且我们看到,每次重新训练模型时,它的表现都在提升,”Söderström 指出。 电话会议上的分析师还询问了 Spotify 对 AI 生成音乐的态度。公司解释说,他们允许艺术家和唱片公司在歌曲的元数据中标记这首歌是如何制作的 (比如是否使用了 AI 辅助),但平台依然会严格监管,打击垃圾内容。 来源:https://t.co/P3QTiUukjn

118
14
25
37.0K
Data updated 1d ago
Viral Probability
67%
Predicted Views
79.0K
Est. 200 views for your reply
Justin
30.2Kfo
Justin@interjc· 1d ago发布

这么看高质量代码一哥又到了 Google 家了 https://t.co/VSjhFteJXO

33
1
14
13.1K
Data updated 1d ago
Viral Probability
52%
Predicted Views
18.0K
Est. 100 views for your reply
宝玉
181.1Kfo
宝玉@dotey· 1d ago发布

https://t.co/PclYVyw7Xc

193
37
8
18.6K
Data updated 1d ago
Viral Probability
79%
Predicted Views
32.0K
Est. 100 views for your reply
Guohao Li 🐫
12.1Kfo
Guohao Li 🐫@guohao_li· 1d ago发布

We definately need more open realistic real-world benchmarks and environments. Great initiative!

17
1
1
1.3K
Data updated 1d ago
Viral Probability
49%
Predicted Views
3.0K
Est. 200 views for your reply
宝玉
181.1Kfo
宝玉@dotey· 1d ago发布

更新一下:我之前把段子当真了,刚用 Codex 分析了 Claude Code 客户端代码(混淆后但是还是明文),没有发现任何 Claude Code 会根据模型更新 Prompt 的证据。 所以我之前说的并不准确,已删帖。 https://t.co/x6ZDKuKRfC

59
3
14
24.4K
Data updated 1d ago
Viral Probability
60%
Predicted Views
38.0K
Est. 100 views for your reply
AI Will
183.9Kfo
AI Will@FinanceYF5· 1d ago发布

根据我目前2.12的认知 快速弄了个Openclaw生态图 一周后迭代! 需要订正请留言 https://t.co/Hz8U9PAIQj

48
13
5
4.3K
Data updated 1d ago
Viral Probability
59%
Predicted Views
9.0K
Est. 100 views for your reply
宝玉
181.1Kfo
宝玉@dotey· 1d ago发布

OpenAI 发布了 GPT-5.3-Codex-Spark,专为实时编程设计的小模型,也是 OpenAI 和 Cerebras 合作后的第一个成果。跑在 Cerebras 晶圆级芯片上,推理速度超过每秒 1000 个 token。https://t.co/oJ6leI4sbb Codex 之前的强项是长时间自主运行,连续工作几小时甚至几天。但日常写代码更多是改个函数、调个接口、重构一段逻辑,等模型想十几分钟再出结果,体验很差。 Codex-Spark 填的就是这个空缺:你可以一边看它输出一边打断、纠正、追问,像跟一个反应极快的搭档对话。 SWE-Bench Pro 上,Codex-Spark 达到 51% 准确率只需 2.3 分钟,GPT-5.3-Codex 同等准确率要 3 分钟,冲到 57% 则需要 16 分钟。 Terminal-Bench 2.0 上 Spark 得分 58.4%,比不上完整版 Codex 的 77.3%,但大幅超过上一代小模型的 46.1%。 OpenAI 顺便把整条推理管线做了优化:引入持久化 WebSocket 连接,往返开销降 80%,每 token 额外开销降 30%,首 token 响应减半。 Cerebras 晶圆级引擎负责极低延迟场景,GPU 仍是训练和推理主力,两者可混合使用。 目前 128K 上下文、纯文本、仅 ChatGPT Pro 用户研究预览。 后续规划是让实时交互和长线任务两种模式融合:Codex 在跟你实时对话的同时,把耗时任务分派给后台子智能体,用户不需要预先选模式。 模型越强,交互速度越是瓶颈,Codex-Spark 是 OpenAI 在这条路上的第一步。

78
8
14
19.2K
Data updated 1d ago
Viral Probability
63%
Predicted Views
28.0K
Est. 100 views for your reply
歸藏(guizang.ai)
124.3Kfo
歸藏(guizang.ai)@op7418· 1d ago发布

Sam 说的今晚专门给 Pro 会员发布的东西原来是: GPT‑5.3‑Codex‑Spark,这玩意每秒能输出 1000+ token。 是 GPT‑5.3‑Codex 的小型变体,专门针对“实时协作写代码”进行优化。 还做了其他优化:每次往返开销降低约 80%、每 token 开销降低 30%、首 token 时间缩短 50%。 https://t.co/BNGwLfXDYu

115
5
11
32.9K
Data updated 1d ago
Viral Probability
57%
Predicted Views
55.0K
Est. 200 views for your reply
宝玉
181.1Kfo
宝玉@dotey· 1d ago发布

https://t.co/wHwjWhDVsc

59
15
3
10.3K
Data updated 1d ago
Viral Probability
64%
Predicted Views
14.0K
Est. 400 views for your reply
歸藏(guizang.ai)
124.3Kfo
歸藏(guizang.ai)@op7418· 1d ago发布

没想到 github 都有这种闲的蛋疼的人,抠字眼非说我这个不能叫原生 桌面 GUI https://t.co/3Rh4KYvllR

33
0
47
25.9K
Data updated 1d ago
Viral Probability
59%
Predicted Views
39.0K
Est. 100 views for your reply
Lexi 勒西
12.5Kfo
Lexi 勒西@lexi_labs· 1d ago发布

OpenAI 的 Atlas 浏览器,我建议大家好好用用。 就说 OpenClaw 用 tailscale 远程访问,clawdbot 出来当天,在Atlas 的 agent 模式帮助下,我也就二十分钟就搞定了。 可我还是把这个功能禁用了。为什么呢? 因为 Atlas 最后说了一句,这个功能有肥尾风险:tailscale 哪天一个漏洞,你可能就倾家荡产了。clawdbot 权限无穷大,风险无穷大,而远程操作的收益,微乎其微,完全不对称。 我深以为然。

42
7
2
12.0K
Data updated 1d ago
Viral Probability
53%
Predicted Views
16.0K
Est. 400 views for your reply
dontbesilent
68.6Kfo
dontbesilent@dontbesilent12· 1d ago发布

四个月之前,我发现哈耶克继承了一些维特根斯坦的思想,发现奥派经济和逻辑哲学,可以合在一起解决商业问题 三个月之前,我用 AI 做了那个“不想说话的中年人”账号,是解决了文字产出的意识形态问题 今天用 claude code 从 0 开始迭代规则,模仿我的文字风格,还没成功,但看见一点希望 上面这三个东西合起来,可以让 AI 产出一段文字,在认知深度、意识形态、哲学坐标、行文风格、排版风格上,都与我相符 甚至应该说,不是复制我,而是实现 “我想实现却未能完美践行的那个版本” 如果跑通了,可能也就是这几天的事儿(但是我还得拍视频、打台球,也可能耽搁到年后) 我预计最终的结果是,可以代替我某些板块的产出,比如日常碎碎念肯定是代替不了(但不会发在我自己的账号,防止污染,会发新号 如果成功了,我能做出一个大家没见过的形态 你可能会看见小红书无数账号开始疯狂“搬运”我,而被搬运的那些东西,都不是我写的,却看起来像是我写的,甚至还能让大家有所收获、更有收获 这是一种不同于以往的矩阵,每个搬运号都在“原创” 不要用陈旧的 “AI 内容代替了人味” 这种思路来理解这件事情,什么什么看不见真人了之类的陈词滥调 所谓 “AI 味”是最最最基础的东西,在我这不存在 我解决的问题是:让合适的角色干合适的事,至于这个角色是不是人,我没有执念

55
1
4
4.7K
Data updated 1d ago
Viral Probability
55%
Predicted Views
10.0K
Est. 500 views for your reply
歸藏(guizang.ai)
124.3Kfo
歸藏(guizang.ai)@op7418· 1d ago发布

谷歌新发布的 Gemini 3 Deep Think 模式确实牛皮 试着让它分析了一下我的 Codepilot 项目,在各个维度和领域都给出了相当深度而且必要的优化建议 比如 Next.js 和进程优化、极限情况下的性能优化以及安全问题、甚至还有交互体验优化 https://t.co/GZklZ54TMZ

90
7
11
38.1K
Data updated 1d ago
Viral Probability
58%
Predicted Views
71.0K
Est. 300 views for your reply
CuiMao
28.8Kfo
CuiMao@CuiMao· 1d ago发布

2026 年的 AI 模型的基准测试和打开鲁大师跑分没有任何区别。

69
1
10
15.8K
Data updated 1d ago
Viral Probability
53%
Predicted Views
21.0K
Est. 100 views for your reply
Wey Gu 古思为
15.1Kfo
Wey Gu 古思为@wey_gu· 1d ago发布

anthropic 太坏了吧…

22
1
5
8.0K
Data updated 1d ago
Viral Probability
46%
Predicted Views
11.0K
Est. 100 views for your reply
歸藏(guizang.ai)
124.3Kfo
歸藏(guizang.ai)@op7418· 1d ago发布

谷歌基于 Gemini 3 的 Deep Think 模式发布了。 各项指标再创新高,尤其是那个 ARC AGI 第二版,比 Gemini 3 Pro 翻了一倍还多。 目前,Ultra 用户可以在 Gemini APP 使用。 https://t.co/5Ietq3Pmy4

67
4
12
55.9K
Data updated 1d ago
Viral Probability
59%
Predicted Views
127.0K
Est. 500 views for your reply
Guohao Li 🐫
12.1Kfo
Guohao Li 🐫@guohao_li· 1d ago发布

80.2% on SWE-Bench Verified and 76.3% on BrowseComp is quite impressive. Try @MiniMax_AI M2.5 on @Eigent_AI https://t.co/R82WRFoh41

23
6
1
3.5K
Data updated 1d ago
Viral Probability
50%
Predicted Views
8.0K
Est. 400 views for your reply
dontbesilent
68.6Kfo
dontbesilent@dontbesilent12· 1d ago发布

我在尝试让 AI 完全从 0 开始模仿我写文字,我不断给 AI 负反馈,让 AI 通过整理这些负反馈,来建立一个写作规则 直到我完全无法提出任何负反馈,那这个模仿的规则就算是建立完成了 那么显然,在这个过程当中,被训练的不是 AI,是我 因为我无法清晰表述我到底是怎么写字的,所以要通过这样的方式去整理规则 AI 帮我发现了很多,我没有意识到的细节(可以进一步增强元语言意识 比如:根本不存在「滤镜之下」的客观对象 我会改成:「滤镜之下」的客观对象根本不存在 把强调的词放到句尾,可以增强语气和冲击力 比如:传统的认识论框架是:XXX,在这个框架下,任务是 YYY 这种句式可以把定义和解释分成两段,让读者有空间停顿和思考 比如:讨论常见观点的时候,经常用「所谓的」这三个字,来保持批判性立场 当这些隐性知识被显性化之后,文字才有了明确的规则,也就有机会去无限复制了

53
3
9
4.7K
Data updated 1d ago
Viral Probability
61%
Predicted Views
10.0K
Est. 100 views for your reply
马东锡 NLP
36.3Kfo
马东锡 NLP@dongxi_nlp· 1d ago发布

MiniMax M2.5 来了! 媲美 Claude Opus 的模型,MiniMax M2.5 端到端的能力非常出色。 “ 1 万美金可以让 4 个 Agent 连续工作一年,M2.5 提供了经济上几乎无限制地构建和运营 Agent 的可能性。” 我非常喜欢 MiniMax 的独特气质,它是我唯一的国产 AI 伙伴。 https://t.co/IQS3MimtYy

110
12
9
29.2K
Data updated 1d ago
Viral Probability
57%
Predicted Views
41.0K
Est. 100 views for your reply
Xiangyu 香鱼🐬
22.4Kfo
Xiangyu 香鱼🐬@XianyuLi· 1d ago发布

写的太好了! 忍不住转发朋友圈! 终于有时间让我缓缓 来看看这篇雄文了!

111
20
5
100.6K
Data updated 1d ago
Viral Probability
100%
Predicted Views
331.0K
Est. 3.8K views for your reply
Ding
144.1Kfo
Ding@dingyi· 1d ago发布

「电子泔水」这个词太精准了,抖音让全民染上了电子鸦片,现在 seedance 又开始制造无数的电子泔水了,最近刷视频已经屏蔽不过来了,全是 AI 生成的劣质垃圾视频。

81
4
12
7.6K
Data updated 1d ago
Viral Probability
63%
Predicted Views
10.0K
Est. 100 views for your reply