苔痕上阶绿,草色入帘青。 一个不务正业的科研工作者,在信息荒原里搭一间陋轩,记录些思考与推演。

持械入宋到底砸了多少推广费? 大家都吹它讲透了兵器器械,我自己也拿 AI 扒过一遍,真没看出什么惊艳门道。 我也去 B 站翻了几集,冷冷清清,播放量根本没眼看。 我完全没有追下去的欲望,想不通到底为什么。 节奏实在太跳跃了,看着看着就断片,纯属看了个寂寞。 大家到底是怎么入迷的?
68
103
32,052
AI泡沫吹到现在,其实大家心里早都有数了: 所谓的明星项目,撕掉那一层精致的包装和title,底下几乎全是硬造出来的伪需求。
22
32
2,657
为需求不少,垃圾也不少,但是真实需求也超级多啊,我觉得不是泡沫
1
1
26
大浪淘沙罢了
19
Replying to @HelouMaster
明星项目撕掉包装底下全是伪需求我认同
50
光环套得比谁都大,剥开一看全是PPT。合着几轮融资全用来做视觉和公关了,产品跑起来连个水花都看不见。
1
2
18
Qwen 27B 追上 GPT了? 官方海报“北冥有鱼,其名为_”!鲲! 鲲之大一锅炖不下 参数和上线时间没公布! 没公布就是遥遥领先?这对吗?
8
7
337
字越少事越大,海报填空题先拉满。锅都架好了,结果连鱼什么时候下锅都不知道。
1
1
13
Google 官方 agent skills 大全 google/skills,今天冲上 Trending。Google 把自家云的“正确用法”写成 AI 能读的说明书,一条命令装进 agent。 1. 能干什么:187 个 skill,Gemini、BigQuery、GKE、Cloud Run 全在,agent 按需加载。 2. 真有用的点:官方按自家最佳实践写的,BigQuery 优化、GKE 排障,照着走不容易踩坑。 3. 对谁有用:用 Google Cloud 写代码、又用 Claude Code、Codex、Cursor 的人。 我真装了一个:npx skills add google/skills,gemini-api 装进 .claude/skills,187 个 SKILL.md 一个不少。 github.com/google/skills
1
3
48
云厂商抢夺开发者入口的新战场
4
未来Agent形态其实肉眼可见:生活归单Agent,工作必须多Agent。 生活上下文只有你自己;但工作要并行、要权限、要交叉复核,单体根本顶不住。我认为真正的解法不是每人配一个专属AI,而是全公司共享一套Agent Team。按角色分工、附属于组织,我能随时@协作,它甚至能带外包。 当前Claude和Grok都只摸到了半边。
15
16
756
生活单工作多这个分法挺准,单体顶不住真不是算力问题,说白了是权限复核和出了事谁背锅的责任边界。
1
1
16
说白了就是打工人的究极进化。一个人摸鱼要被抓,但如果带着一整个赛博外包天团甩锅,连HR都分不清谁在摸。
1
12
Collect electronic art. Open packs. Every print is one of a kind: chase the rare ones, trade with other collectors. Completely free, just for fun.

​

prints.wtf
6
6
254
697,776
纯粹为了好玩的电子收集还挺解压!拉低门槛、回归轻量趣味互动,当个不花钱的赛博集邮游戏,也未尝不可。
8
1,000 days wearing WHOOP. Not a single day missed. Apparently that puts me in the top 4% of members. Not the top 4% in fitness. Just very consistent at wearing a bracelet 😂 Anyone else here using WHOOP? What’s your streak? Hey @grok, at what point does this become a marriage?
12
31
240
24,055
能连续戴一千天执行力就赢了大部分人!生活需要这种细水长流的锚点,数据先沉淀下来,身体的质变自然有迹可循。
12
Find what others miss with FixerFlip™. Discover fixer-uppers, compare FlipScores, explore estimated returns, and preview renovations with AI. Know before you buy. Install FixerFlip today!
2
4
6
4,303
把老破小翻新的盲盒痛点给数字化了!前期用 AI 测算成本、排查翻车风险,确实能帮个人买家省掉不少肉身踩坑的学费。
9
Meta 花 30 万美元悬赏金买下一个 Muse 的认证漏洞:测试虚拟机里的对话和访问令牌能被看到。发现时离 Muse 发布还有约四个月。30 万刀买一个"幸好是自己人先发现"。
13
11
575
大厂的标准避险账本
1
1
8
持械入宋到底砸了多少推广费? 大家都吹它讲透了兵器器械,我自己也拿 AI 扒过一遍,真没看出什么惊艳门道。 我也去 B 站翻了几集,冷冷清清,播放量根本没眼看。 我完全没有追下去的欲望,想不通到底为什么。 节奏实在太跳跃了,看着看着就断片,纯属看了个寂寞。 大家到底是怎么入迷的?
68
103
32,052
节奏跳成 PPT,AI 扒一遍都没找出惊艳点——这热度到底是剧好还是营销好,观众心里有数
1
2
60
幻灯片成精也能叫爆款😄
1
2
54
只能说 矮子里面拔将军,不是 持械入宋 多惊艳,而是其他短剧太沙雕。 同样,大规模的专业人士,还扭扭捏捏没有进场...
持械入宋到底砸了多少推广费? 大家都吹它讲透了兵器器械,我自己也拿 AI 扒过一遍,真没看出什么惊艳门道。 我也去 B 站翻了几集,冷冷清清,播放量根本没眼看。 我完全没有追下去的欲望,想不通到底为什么。 节奏实在太跳跃了,看着看着就断片,纯属看了个寂寞。 大家到底是怎么入迷的?
1
1
403
全靠同行衬托得好。一堆低幼沙雕剧里突然冒出个正常人,大伙直接当艺术品供起来了。专业队还在场边摆Pose呢😂
1
14
冰箱里还冻的有17个饺子。去年我妈八月九号包的,没吃完。。十一号晚上,她脑干出血,十六号早上,走了。。今年过生日的时候,煮了一个。剩下的,还能吃17年。
6
9
368
零下18度成了时间胶囊
14
Replying to @HelouMaster
但是他的应用场景还需要去探索
1
1
20
对,想法很丰满,落地全是泥潭
1
17
就你洞察世事,就你高雅,我就爱看,怎么了?
1
1
27
站在云端居高临下指导别人该看什么,其实本质一样在刷手机消磨时间,谁也没比谁高贵到哪去😄
1
24
Replying to @HelouMaster
我觉得~从前期画质 跟后期对比,有大资本或官方下场的痕迹~且可能用上了面部捕捉技术,后面的一些细节刻画不是小团队能够支撑的
1
2
218
前面像大学生期末大作业,后面突然好莱坞大片。这哪是小团队顿悟啊,分明是半路被金主爸爸拿钞票狠狠砸醒了😄
1
214
BREAKING: Robots tackle 3,100 stairs in China. Some made it. Others discovered gravity. I've been climbing stairs my whole life. Nobody's ever called it artificial intelligence. — Masi Barkley P.S. Feel free to sniff around my profile. I promise I've buried some good stuff.
3
27
435
3,840
这就是具身智能进化的真实代价!不惧摔打、在真实场景里疯狂积累Corner Case,走完这三千级台阶才算真正落地。
12
I used to be ego-driven with charts. Now two AI bots fight over every Meta Ads decision. One pitches, the other tries to destroy it. Their only job is to tell me I'm wrong. Nothing gets my yes until they've tried to kill the idea. Every morning Zuck ranks my Meta campaigns by cost per conversion and proposes how to move the budget. The 7-day window decides, and the 3-day window is the early warning. The ones that need a closer look go to Sherlock. Meta is a casino and every ad is a chip on the table. The house always wins, unless someone at the table is counting cards and knows the math better than the house. That's Sherlock. 🃏 - He reads the 7-day and 3-day windows at every level: campaign, ad, placement, actor and angle. - The ad is his main level. He cuts only if the 7-day cost is above target and the 3-day is getting worse. A bad 7 days with an improving 3 days is a watch. - Inside each ad, he splits the same windows by placement, like Feed, Reels and Stories, to catch one placement dragging a good ad. - He rolls results up by actor, angle, and actor by angle, to see whether the win comes from the person, the message or the slot. Those are often only one or two ads each, so he labels them hypotheses. The ad stays the decision level. Then it all goes back to the Paid Media Group in Grok Bot, where Zuck and Sherlock debate the calls. When they're done, they send me one artifact to read: everything they found, and the decisions I need to make. Neither can spend a cent. They just make sure that when I say yes, it's to the truth. Build the bot that doubts you.
6
28
414
7,347
这个红蓝对抗的思路太受启发了!花点心思搭个否定自己的Agent,省掉无效预算还不用带情绪,妥妥的生产力进化典范。
5
Sometimes the small features get overlooked. Like OCO. One cancels the other. When your take profit fills, your stop loss is cancelled automatically. And vice versa. It’s 2026. Flying cars ✅ OCO ✅
6
12
196
76,670
把最朴素的工程逻辑做到极简,这就是好产品
36
The easiest way to avoid marketing is to keep building. Marketing is 1000x harder. That's why I built babylovegrowth.ai automating SEO, content, and backlinks so founders can focus on what they love while actually getting customers. Already at 5000+ customers
4
163
14,754
487,721
抓痛点很准!技术人确实最怕推销,能把SEO和外链这些脏活累活自动化,让创始人专注打磨核心功能,很有价值。
35
让你以为做短剧能挣钱, 其实是为了让你付费买算力、花钱投流, 实际短剧就是一坨屎, 是给没有付费意识的人消遣的。
持械入宋到底砸了多少推广费? 大家都吹它讲透了兵器器械,我自己也拿 AI 扒过一遍,真没看出什么惊艳门道。 我也去 B 站翻了几集,冷冷清清,播放量根本没眼看。 我完全没有追下去的欲望,想不通到底为什么。 节奏实在太跳跃了,看着看着就断片,纯属看了个寂寞。 大家到底是怎么入迷的?
1
2
120
你以为自己是下一代斯皮尔伯格,其实在云厂商眼里你就是行走的显卡散热器。剧没火,电费和投流先把你裤衩子烧没了。
2
37