Masterdai (@MasterXDai) header
Masterdai (@MasterXDai) profile picture

I built AI wrapper -10i $ MRR

Joined January 2014
Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)Photo by Masterdai (@MasterXDai)
感觉是不同部门之间的矛盾。或者说就没考虑过会影响。
Google shoved AI summaries down everyone’s throats, only to realize people were clicking fewer website links because AI was giving them the answers. Now it’s costing them billions in lost ad revenue. Incredible that anyone with half a brain couldn’t have seen this coming.
54
我其实对于田柯宇的创业内容不感兴趣,我之前倒是对他的作案动机十分感兴趣。 因为我第一次听到大模型投毒,还以为他是一个类似泰德卡辛斯基这样的人。他可能有一些宏大的、自洽乃至离谱的逻辑,去支撑他实施这样的违法行为。 比如说,他觉得 AI 发展得太快,剥夺了普通人的生存空间,或者预见了未来只有少数人掌握技术、其他人被永久甩在下面的反乌托邦社会,从而决定从源头掐灭这个万恶之源。 当然,我也不是觉得动机宏大,违法行为就变得高尚了。我感兴趣的是,他到底反对什么、预见了什么,又是怎么说服自己,觉得自己有权去阻止这样的未来。 但是听完播客之后,大失所望。至少按他自己的说法,无非就是别人先欺负人,领导不管,我看不下去,所以替大家出头。 我发现他的动机和行为,简直就是这片土地里面反复发生的一个个故事的翻版。 你把训练资源、投毒大模型、植入代码,替换成给村里的集体鱼塘投毒、撕掉学校里的课题集、争公司里面的奖金,都不违和。 我失望的倒不是他的动机不够崇高,而是这件事居然没有多少思想上的独特性。无非就是里面的人学历高了点,手里的工具先进了点罢了。 当然,我还是觉得他不会讲故事。倒不是替人出头这个动机太普通,优秀的文学和影视作品里,不乏带着正义的口号,最终酿出悲剧的故事。 关键是,他有没有想过自己的正义凭什么成立,又是怎么变成了越界的理由。 但到了牢田嘴里,还是别人欺负人、我替人出头、只是方法不太成熟,最后讲出了以暴制暴、反所谓霸凌的网络小说既视感。
235
Masterdai retweeted
Here is how to enable a 1M-token context window in Codex for GPT-5.6 Sol. Even though we have tuned the context limit in Codex to be set optimally when it comes to performance and cost, this is a common ask, so here it is documented. A larger context window lets Codex retain more code, tool output, and conversation history before summarizing older material. You need a model that supports it. And GPT-5.6 Sol, for example, has a documented 1,050,000-token window. Open ~/.codex/config.toml and add or update these settings at the top level, before any [section] headers: ``` model = "gpt-5.6-sol" model_context_window = 1000000 model_auto_compact_token_limit = 900000 ``` The first setting selects the model. The second tells Codex to use a one-million-token context budget. The third starts automatic history compaction around 900,000 tokens, leaving some headroom. Restart Codex client and start a new session after saving. To try the configuration for a single CLI session without changing your defaults: ``` codex -m gpt-5.6-sol \ -c model_context_window=1000000 \ -c model_auto_compact_token_limit=900000 ``` Have fun, but also know that we tuned the default carefully!
1,007
1,175
543
14,861
3,850,921
对的,所以我养成了下意识点赞收藏的习惯
Number one thing I hate about X That half second where an interesting post flashes by, you go “wait, what was that,” and the timeline refreshes it into oblivion
1
1
225
Anthropic onboarding day: Michael Scott introducing Karpathy like he just signed Wemby in free agency.
389
1,465
375
17,588
2,482,110
Masterdai retweeted
在研发 Evot 过程中发现,要让 Anthropic(Opus 系列)模型发挥到极致,官方 Claude Code 的做法基本是最优解,很难绕过。 对 Claude Code prompt 做了深度分析和量化验证,发现他们在训练阶段就把 Agent Harness 的行为模式烧进去了: 1. 长内容输出时模型会自发产生 <!-- PLACEHOLDER_ --> 占位符做分块写入——system prompt 里压根没要求这么做 2. Tool name 在模型内部有偏好绑定,只有官方名字效果最佳,换个名字行为差异巨大 ... ... 这不是 prompt engineering 的范畴了——模型训练时就把 scaffold 的行为模式编进了权重。第三方想达到同样效果,要么对齐官方,要么自己训。。 猜测 DeepSeek 搞 Agent Harness(@victor207755822),也得走同样的路——把 harness 行为烧进模型训练,模型和 scaffold 深度耦合。Cursor Composer 2.5 大概率也是这么干的,1/10 成本追平 Opus。 Agent Harness 下一个竞争维度:agent 行为从 prompt 层下沉到模型层。prompt 和权重里外配合,效果远超纯靠 prompt 硬拗。
13
20
7
137
27,232
Masterdai retweeted
很多人对第二条没有很深刻的认知。 成本,资本对公司发展的约束。 事实上很多人在一个快速增长的创业公司挣到的钱远远比自己创业挣到的钱多。 为什么呢? 资本金规模的硬约束。 创业/自己做生意,要看你的资本金的来源。如果你一开始就是从亲友,从自由资金,从一些小资本拿到钱,假设你凑了100万美元,你占股60%,你项目很好,执行很得力,你每年增值的速度是YoY是50%,那么你做5年的你的规模差不多能做到760万的规模。 规模被这个初始的100万给约束住了 突破这约束的一个办法就是想办法融更多的资,把这个资本金规模做大。 假设在在这五年内资本融到了1个亿,创始团队的股权降到了10%。 然后整个公司的规模达到了3个亿。。 那么即便虽然创始团队的股权被稀释了,但是他们股权远远超过了原来的价值。 所以很多创业公司的老板一个非常非常重要的任务,就是给各路资本画饼,融资。 但是能够持续融资是极少数公司,是资本的宠儿。大部分创业者是做不到的。做不到能持续融资,做不到能保持yoy 50%的增速。 所以你会看到大量的创业者公司做了十几年、规模一直原地踏步。 这次AI 浪潮,硅谷大概创造了几万个净资产超过2000万美元的超级打工人。 他们并不一定比其他行业的人更努力,更聪明,而是他们在的公司,是资本的宠儿,他们的老板能够融到天量的资本,他们的公司能维持高增长。 This is how capitalism works. 所以很多优秀的创业者很早就想好了公司的发展途径和融资的计划。 没有对错,都是个人的选择。
看来大家对我昨天写这篇应届毕业生怎么做OPC很关注啊 虽然这是我第三次创业,我第一次东莞创建了服装厂,第二次做Deepin Linux,第三次做懒猫微服这个公司 但是我的建议是,应届毕业生依然不要去创业。不到迫不得已,不要去创业 从表面上看,大家说创业的逻辑是风口啊,是你的能力呀,是你产品的价值呀 但是从我将近20年的经验来分享啊,创业其实最后考验的是创始人的心态 1. 兜底的心态,因为你原来跟别人打工的时候,你可以觉得这里不好,那里不好。你可以抱怨,你可以吐槽,你可以让上级帮你忙。但是你真正有一天作为老板以后,你才知道,做公司就是一堆屁事,其实你内心一点都不想管,你只想把销售搞上去。当你真的有一天自己创业的时候,你会发现,那种自己跟自己兜底的孤独感和无助感,这是很多人心理很难跨越的事情 2. 对成本的控制,很多没有创业的人,就是说,哎呀,我有很牛逼的想法,我有很多的灵感,然后我能干翻全世界。不管你是多么强的团队,你在重新创业的时候,你的产品、你的组织还有你的品牌都是新的。你从有经验的创业者到最后能成功运营一家企业,都要面临同一个问题,他是需要花3到10年时间去成长的。不管你是年轻的创业者还是成熟的创业者,坚持3到10年,其实就是要对成本控制非常敏感。什么意思呢?就是有多少能力、有多少资金做多少事情,不要做超过你能力和资金的事情,可以慢慢做。一旦你上了面子,一旦你要跟别人争强好胜,你就会忘记成本的控制,你就会死得很快 3. 最后才是热爱,客观的跟你们讲,其实我做公司的时候,每天一睁眼就一堆破事。还要教团队,还要培训团队。然后团队遇到问题了,会把最难啃的骨头、最硬的玻璃都倒给我。说句实话,有时候挺困难的,心里很难受,想放弃。但是真正能让我坚持下来的,其实就是我热爱这个事情。比如说我就喜欢玩Linux,就喜欢折腾这些底层细节,就喜欢折腾完以后把它变成产品化去服务用户。因为我看到用户用我用心打造的产品,节省他们时间,他们很快乐,他们说一句感谢,我觉得这就是我的追求循环吧 好了,说到最后,大多数年轻人,你们的精力非常旺盛,你们有对未来无限的憧憬。但是只有很少很少很少的人能做到创业者的心态,因为创业者就是一个从地狱走一遭的磨砺。如果你自己不是一个内心非常坚强的人,我是非常非常不建议毕业就开始创业,这是非常不理性的事情 当然,如果你依然坚持创业,我祝你好运。因为创业是世间最好的修道场,最好的入世修行的方法 好了。今天的分享到这里,喜欢我创业故事和创业经验的朋友记得点赞、收藏、转发 好奇我们团队做的事情,可以点开我个人主页的置顶贴,欢迎采购懒猫微服。懒猫微服是AI时代的编程利器,可以极大提升你的工作效率。评论区打1,我还可以给你专属的优惠
6
21
3
182
81,902
最近 seedance 2.0 api 和 anthropic muthos 等对普通用户的限制。 让我有一种强烈的感觉:过去两三年的 AI,实际上是一段田园牧歌式的开放期,而我们正在亲眼看着它走向封闭。注册一个账号就能用上世界最强的模型,这种体验在未来回看,可能真的是一段不可复现的黄金时期。
1
304
AI CEO vs. Engineer (2026).
31
189
36
1,505
135,133
Tobi Lutke explains what the VCs who passed on Shopify got wrong Tobi recounts pitching Shopify to VCs on Sand Hill Road a few years after founding Shopify. Investors passed because they thought the addressable market was too small. At the time, there were about 40,000-50,000 online stores, and even if Shopify captured 50% of the market, that still wouldn’t be a venture-scale business. When Tobi ran into the VC partner a few years ago, the partner asked Tobi what he missed (Shopify is valued at almost $100 billion today). Tobi explained: “You were actually correct, but what you didn’t realize was that Shopify was the solution to the very problem you identified. The reason there was only 40,000 online stores was because it was hard, expensive, and everyone who tried ran into all these brick walls of complexity, which Shopify, one after another, smoothed over and made simple to do.” Tobi believes this is a common mistake: “What a lot of free-market thinkers don’t understand is that between the demand and eventual supply lies friction. And I actually think that friction is probably the most potent force for shaping the planet that people just generally do not acknowledge… That was my theory when I turned my snowboard store into Shopify: there was a lot more people like me except there was too much friction which we needed to solve. And Shopify has proven out that every time we make the process simpler, there’s more consumption. At this point, we have a million merchants on Shopify, which is a mind-blowing number. So friction is a major component, and it’s something that software is uniquely good at reducing.” Video source: @danmartell (2019)
37
75
32
952
169,671
Masterdai retweeted
We've withdrawn the model. Just use Nano Banana Pro with "Improve Prompt" button for similar results and the same price. Happy to give free credits to anyone who wants to try.
After further testing of @PhotaLabs' Image "Model", we have concluded that Phota is merely a Nano Banana Wrapper that is being sold as a 'new foundation model'. This is deceptive and doesn't align with our values, nor our value proposition. We have therefore decided to withdraw this model again, after having added it earlier today. We apologize for any confusion caused, and ask PhotaLabs to adjust their communication.
7
2
2
108
19,522
我现在站在自己创业的视角去看待很多人公开和非公开场合说的话,我能分辨出来 哪些是 画大饼的 哪些是为了安抚团队的。 我学到宝贵的经验就是 真正的信念是用放弃来证明的,不是用宣言来证明的。 你选择做什么不重要,你选择不做什么才暴露你真正相信的东西。 所以当我听到别人说大话的时候,我会本能地去看: 他说这句话之后,下一步做了什么? 我倒是希望自己“诚实”一点 或者说知行合一 因为我知道 有一天我的员工、我的合作伙伴也会用同样的方式审视我。
1
126
Masterdai retweeted
Replying to @thulynnn
我觉得不是“假” 是不够诚实,我看到这段话的时候,我就已经开始怀疑了。 因为他的愿景和其真实的行动是背道而驰的。 一个人嘴上说着最大的愿景,手上做着完全相反的选择,只有两种可能: 他在骗别人——明知道要走,先稳住团队 他在骗自己——从未严肃思考过那句话意味着什么代价
1
1
115
"身不由己"这个叙事最大的问题是:它取消了行动者的主体性。把一个创始人、一个融了七千五百万美金的团队,描述成时代洪流中的落叶——这本身就是一种不诚实。 真实的链条是: 选择接受美国资本 → 选择依赖闭源模型 → 选择迁往新加坡 每一步都是主动行动,每一步都排除了其他可能性。说"被逼上绝路",但没有人逼他们拿Benchmark的term sheet签字。 出海是选择,留下也是选择。选美国资本是选择,选国内资本也是选择。每一条路都有代价,但承认自己选了利益更大的那条路,远比装作被命运推着走要诚实得多。 阿伦特说人是自由的,潜台词是——你不能事后假装自己不自由。
大家与其看所谓的 manus 虚拟故事,不如我给大家讲讲真实的事情发展。 作为前员工和做过传统行业,和国央企,市区县一级政府深度打过交道的人。大家对两者都有不同程度的误解,这些误解有些来自于信息的缺失,有些则完全是偏见和想象。我只能说这场风波里 manus 和中国政府都是不完美的受害者。 很多人已经忘记了,manus 第一次面临来次政府的压力,不是中国而是美国。25年 3月,manus 爆火,4月末,benchmark 领投 $7500万,转瞬五月中,美国财政部就问询 benchmark,内容是这笔投资是否属于针对“关注国家”先进技术(如人工智能)投资的新限制范围。 那个时刻,大模型,确实只有anthropic,openai,gemini。开源模型一个能打的都没有。美国同时又是资金最充沛,用户付费意愿最高的市场。可在中美脱轨背景下,美国不太能容下中国公司拿着美国人的投资挣美国人的钱。 很多人不知道,就在 benchmark 投资后,美国问询前,内部有过一次 all hands 大会,那时候 red 很骄傲地说,他的愿景是做一个被世界认可的“中国公司”;没错,很难想一个大家现在快被黑成什么的人,其实抱着赤子之心的。 但现实真的很难。那次问询几乎把 manus 逼上绝路。拒绝美国,不仅是拒绝整个美国的资本和市场,还等同于让当时绝对依赖闭源 sota 模型的 manus 在产品上自杀。而 manus 的成功已经就是肉眼可见的,那种具体清晰的果实给人的诱惑,无法抗拒。 而在国内,3月的全民欢呼,已经让 manus 和某种情绪有了联系。manus 的Logo出现在各种官媒上,甚至是CCTV,稍微有点政治敏感度,都会知道这背后,是来自官方的认可,而认可背后,就代表这,manus的背离要打这些所有人的脸。一根筋变两头堵。 世界是如此荒谬,manus 每一次的困境,都是因为成功造成的。 于是就有了七月份搬到新加坡的决策。这个决策怎么形成的,无从得知。但没有发生所谓,把中国团队直接裁掉,当时的策略是,在EP(工作签)可以满足的情况下,能够去新加坡都去。不能去的,顶格赔偿。除了搬去新加坡这件事情上,对于员工的处理无可指责。 而且在问询前,团队定的基调是要迅速扩张,七月份走的人,绝大部分,是manus三月份火了之后,招进来,这批员工,不仅拿到大礼包,而且因为当时正值 manus 如日当天,很多人都靠这段经历,转身就进了大厂。
2
1
466
OpenAI shut down Sora. This was inevitable — and it reveals a fundamental truth about building AI products that most people still don't get. In the traditional internet era, free worked beautifully. The marginal cost of serving one more user was basically zero. More users meant better unit economics. You could give the product away for free, monetize through ads or upsells, and print money at scale. That playbook built Google, Facebook, YouTube, and almost every major consumer internet company. But generative AI breaks this model completely. Every single inference call burns real GPU compute. When Sora generates a 10-second video, it costs roughly $1.3. That cost doesn't go down when you add more users — it goes up. Linearly at best, super-linearly at worst when you factor in GPU scarcity and peak-hour demand. Forbes estimated that with ~4.5 million users, assuming just a quarter of them generated 10 videos a day, Sora was burning through approximately $15 million per day in compute costs. That's over $5.4 billion annualized — for a single product — while OpenAI was already losing $12 billion a quarter. The math was never going to work. And now Sora is dead. Meanwhile, look at what Chinese AI companies like Kuaishou (Kling) and ByteDance (Seedance) have done. They figured out something OpenAI didn't: in the AI era, the free strategy isn't dead, but you have to be extremely intentional about what you make free. Image generation costs about $0.03–0.04 per generation. Video generation costs $1+ per generation. That's a 30-40x difference. So they made image generation free or near-free to hook users and build habit, while charging for video generation from day one. Free where you can afford it, paid where the costs demand it. This is the correct framework. OpenAI did the opposite — they gave away the most expensive thing for free and hoped the economics would somehow work out later. It's like a restaurant giving away the steak for free and charging for the bread. The deeper lesson: AI products cannot blindly copy the internet-era growth playbook. The "get big fast, monetize later" approach assumes near-zero marginal costs. When your marginal cost is $1.3 per user action, getting big fast just means bleeding out faster. The companies that will win in AI are the ones who understand cost-per-inference at a granular level and design their entire product and pricing strategy around it — not the ones who raise the most money and burn it the fastest. Sora's shutdown is the first major proof point. It won't be the last. forbes.com/sites/phoebeliu/2…
132
best sora video i made ever
We’re saying goodbye to the Sora app. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing. We’ll share more soon, including timelines for the app and API and details on preserving your work. – The Sora Team
116
open ai 在sora 的运营和维护真的就是拍脑袋一样。 我分享一个很细节的东西,就是sora 2当时开发api的时候,我是最早一批用他们的服务搭建流程的。然后当时sora 2的 官方 文档里面写支持用oai的webhook ,但是我找了很久都找不到参数。最后我去论坛求助才发现原来他们忘了加上去😐 最后隔了两天才修好。 sora app的问题就是,这帮人拥有当时最好的视频模型,偏偏要免费给客户用,而且免费的app里面生成视频的速度比api调用的还要快。 给我感觉就是本末倒置。
We’re saying goodbye to the Sora app. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing. We’ll share more soon, including timelines for the app and API and details on preserving your work. – The Sora Team
1
324
我的天呐这个ai 生成的也太好了
Finally here! Dreamina Seedance 2.0 and Seedream 5.0 Lite are Now Available on Dreamina AI We're excited to introduce our latest models, which are built for deeper control and broader creative possibilities. What Makes Dreamina Seedance 2.0 Exceptional? → Multimodal video model → Reference with image, video, and text inputs → Ultimate consistency across multiple scenes → More expressive, more controllable Create images with Seedream 5.0 Lite → Real-time world knowledge → Follows your prompts, delivers precise results → Upgraded visual expression We are gradually rolling out Dreamina Seedance 2.0 and Seedream 5.0 Lite to our community over the coming days. Hang tight if you don't see it yet and check back soon. Try it now: dreamina.capcut.com/ai-tool/… #DreaminaSeedance2 #DreaminaAI #Seedream5Lite #Seedance2
95
这次更新之后 微信的插件不能用了🤣。
OpenClaw 2026.3.22 🦞 🏪 ClawHub plugin marketplace 🤖 MiniMax M2.7, GPT-5.4-mini/nano + per-agent reasoning 💬 /btw side questions 🏖️ OpenShell + SSH sandboxes 🌐 Exa, Tavily, Firecrawl search This release is so big it needs its own table of contents. github.com/openclaw/openclaw…
110
I finally solved the character consistencies problem in seedance2.
60