Masterdai retweeted
show this video to anybody that is an old school programmer that hates AI
this is one of the best videos I’ve seen all month.
Masterdai retweeted
Did it hit you that a centibillionaire, Jensen Huang, is less free than a random plumber?
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!
Masterdai retweeted
Anthropic onboarding day: Michael Scott introducing Karpathy like he just signed Wemby in free agency.
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。
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
Masterdai retweeted
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)
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.
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
Masterdai retweeted
I recently came across a very clear AI KOL monetization case on TikTok.
The account only has around 13.6K followers, but it already has 5.3M likes, and several videos are doing serious numbers
What is interesting is not just the traffic. It is how tightly the content format and the monetization system are connected.
A lot of the content is basically slide-style storytelling + AI hook copy.
It does not open with information.
It opens with emotion:
“For 6 months...”
“I couldn’t sleep..”
“She’s ugly, what’s wrong with her face”
Those lines are not teaching anything. They are triggering a stack of very basic audience instincts at the same time: curiosity, judgment, voyeurism, empathy, and aspiration.
Then the account adds a very strong before/after contrast.
On one side, you get a version of the character that looks heavier, less polished, and more vulnerable.
On the other side, you get a version that looks slimmer, more refined, and fully “glowed up.”
That contrast creates one question in the viewer’s mind:
what did she do to change?
And there is another smart detail here.
If you look at this as an AI KOL system, the creator did not choose a generic pretty face. They gave the character a very exaggerated facial birthmark.
That does two things at once:
1. It makes the character instantly recognizable.
2. It works as a consistency anchor for AI image generation, because the same character can be identified across different outfits, scenes, and slides.
So the birthmark is not just a story element.
It also helps the AI persona stay visually consistent.
The monetization path is also extremely clean.
The bio says:
“I don’t gatekeep”
“Secret to my glow up below”
And the traffic goes to bodyglowup website
That page is basically a low-ticket body transformation funnel packaged as a glow up product. It is selling a roughly $9.30 downloadable routine around waist slimming, glute growth, debloating, and a 30-day body transformation.
So the full system looks like this:
AI KOL persona
-> slide-style emotional content
-> before/after tension
-> repeated viewer question
-> short bio CTA
-> low-ticket body offer
Masterdai retweeted
Replying to @KirkMarple
The only use Product Hunt has for me now is as a decent, high-DR free backlink.
Masterdai retweeted
gary 的问题就在于,他作为一个投资人,几乎很少会听到负面的反馈。尤其是他把这个东西拿给一个“cto 朋友”看,这个 cto 朋友大概率也只是出于客气,给了他一些虚假的称赞,于是他就更有信心,觉得这是一个“伟大的产品”。
但问题在于,不只是 gary。到底有多少人是因为他是 YC CEO,才去 Product Hunt 给他点赞、去 GitHub 给他点 star,而不是因为这个产品真的有用?
反过来,又有多少真正有用、有价值的产品,在这样的平台上得不到关注,只是因为它是一个 nobody 发布的产品?
gstack 争议的重点,不是它是不是 prompt,而是它的热度里到底有多少来自产品本身,又有多少只是来自“这是 YC CEO 发的东西”。
Product Hunt 点赞、GitHub stars、社交媒体喝彩,很多时候衡量的是身份、流量和分发,而不是价值。
Masterdai retweeted
Replying to @atmoio
Gary’s problem is that, as an investor, he rarely hears negative feedback, especially about himself. He showed this to a “CTO friend,” and that friend probably just gave him polite, empty praise. That was enough to make him believe this was a “great product.”
But this is bigger than Gary. How many people upvoted it on Product Hunt or starred it on GitHub just because he is the CEO of YC, not because the product is actually useful?
And how many truly useful products get ignored on these platforms simply because they were launched by a nobody?
Masterdai retweeted
Wow, this is a dynamic special effect made using Kling V3!
Just take a photo, use the Kling V3 model, and enter the prompt below to get the “character in the computer shaking hands with you” effect.
prompt
"Use the uploaded image as the base reference and keep everything exactly the same: the computer monitor, the character on the screen, the lighting, and the real human hand in front of the screen. The scene starts as a still image. After a moment, the character on the computer screen slowly becomes animated and looks toward the human hand in front of the monitor. A subtle digital glow appears on the monitor and the screen surface begins to ripple slightly like liquid glass. The character slowly raises their hand toward the screen from inside the display. The hand begins emerging out of the monitor as if the screen is a soft digital portal. The character’s hand moves forward from the screen toward the real human hand. Both hands meet and shake hands in slow motion. At the moment of contact, a subtle glowing pulse spreads from the handshake point. The handshake looks natural and realistic while the character remains partly inside the screen and partly outside. Keep the original character face and design exactly the same as the input image. Cinematic lighting, dramatic shadows, shallow depth of field, smooth camera movement, ultra realistic motion. 8–10 second cinematic video, photorealistic, smooth animation, emotional cinematic moment. Important: do not change the original image composition, only animate the scene.
"
Masterdai retweeted
I have been testing the new Obsidian CLI with Claude Code on my research vault (4,663 files, 16 GB)... I know too many notes!!
Early results are significant. Its going to change the way in which Claude Code can interact with Obsidian
The way I see it, there are three ways Claude can connect to your vault:
Filesystem (MCP or bash) reads/writes markdown files. Covers maybe 40% of what Obsidian actually knows. No awareness of backlinks, tags, properties, or the graph. To search content, it has to open every file individually.
REST API MCP — talks to Obsidian via plugin. Gets you to about 55%. Better search, some metadata. But fragile setup and limited.
Obsidian CLI ...Yay @obsdmd and @kepano !! it queries Obsidian's actual indexes. This would be about 85% of Obsidian's capabilities. The missing 15% is purely visual the canvas layout, graph view rendering, live preview. Everything else is there: search, backlinks, orphan detection, properties, tags with hierarchy and counts.
The speed difference is real:
Finding orphan notes: bash grep 15.6s vs CLI 0.26s (54× faster)
Searching vault: bash grep 1.95s vs CLI 0.32s (6× faster)
Token cost for orphan detection via MCP: about 7 million tokens. Via CLI: 100 tokens. That's 70,000× cheaper.
The CLI uses Obsidian's pre-built search index the same thing that makes Obsidian's own search instant. Grep scans every file from scratch every time.
The catch: right now this only works via Claude Code (which can run CLI commands through bash). Claude Desktop and claude.ai can't access it directly. There is an early CLI MCP server (obsidian-ts-mcp) that would bridge this gap but I haven't tested it yet. (I think that if you ask nicely Claude Code could create a version for you!)
I'm using this as part of a research assistant stack connecting Claude to Obsidian, Zotero, PubMed and more. Posts on each piece coming.
Masterdai retweeted
OpenClaw creator @steipete sets up OC for his non-technical friend, pays the $100 Claude Max sub, and gets him hooked instantly
He loves it. He's building and learning so many things he upgrades to the $200 Claude plan
Few days later Anthropic blocks his account, devastates him, and causes him to switch providers




























