美国AGI每天都在扔核弹，中国的AGI小米加步枪。
为了中国的AGI继续努力🏃振兴中华🇨🇳

江浙人的春菜是什么人间天堂，2块一个的甜酒酿馒头吃哭了🥲

知行合一，英明不朽，物华天宝，人杰地灵。

会稽山阴兰亭本亭，淡极始知花更艳。

永和九年，岁在癸丑，暮春之初，会于会稽山阴之兰亭，修禊事也。群贤毕至，少长咸集。此地有崇山峻岭，茂林修竹，又有清流激湍，映带左右，引以为流觞曲水，列坐其次。虽无丝竹管弦之盛，一觞一咏，亦足以畅叙幽情。
是日也，天朗气清，惠风和畅。仰观宇宙之大，俯察品类之盛，所以游目骋怀，足以极视听之娱，信可乐也。
夫人之相与，俯仰一世。或取诸怀抱，悟言一室之内；或因寄所托，放浪形骸之外。虽趣舍万殊，静躁不同，当其欣于所遇，暂得于己，快然自足，不知老之将至；及其所之既倦，情随事迁，感慨系之矣。向之所欣，俯仰之间，已为陈迹，犹不能不以之兴怀，况修短随化，终期于尽！古人云："死生亦大矣。"岂不痛哉！
每览昔人兴感之由，若合一契，未尝不临文嗟悼，不能喻之于怀。固知一死生为虚诞，齐彭殇为妄作。后之视今，亦犹今之视昔，悲夫！故列叙时人，录其所述，虽世殊事异，所以兴怀，其致一也。后之览者，亦将有感于斯文。

OpenAI宣布以约2.5亿美元收购科技播客TBPN，并下调ChatGPT订阅价格。Anthropic 用最强模型做增长黑客。感觉前者有点动作变形了。

Claude的200美元Max计划实际提供了价值4000-5000美元的使用量。这不是亏本生意，而是Anthropic的增长黑客策略：那些用到极限的重度用户本质上是产品的免费营销人员——他们会主动向朋友和公司推荐这个产品。所以不要把它看成毛利率问题，而应该看成营销支出。随着品牌知名度提高，这些补贴最终会减少。

真正的理想主义是，当他被时代选中的时候，从乌托邦实验中走出来直面挑战。

👍

Anthropic 的联合创始人 Chris Olah 曾说过，像 Claude 这样的生成式 AI 系统，与其说是被“制造”出来的，不如说是被“培育”出来的。研究人员设定好引导它们生长的条件，但最终会演化出怎样的确切结构或能力，往往是无法准确预测的。这给使用 Claude 开发应用带来了挑战：开发者通常会使用 AI 套壳或智能体框架 (agent harnesses)（指用来包裹、控制和辅助 AI 模型运行的外部代码结构）来弥补 Claude 自身做不到的事情。这些框架基于各种“假设”构建，但随着 Claude 能力的不断进化，这些假设很快就会过时。因此，即使是本文分享的经验之谈，也需要你时常温故而知新。

未来属于：用好Claude的人类

@ Rubén Domínguez Ibar：In 12 months, the people who mastered Claude will be untouchable 🤖
Stock research like a hedge fund analyst www.the-ai-corner.com
How to use AI to Research Stocks like a Hedge Fund Analyst
8 copy-paste prompts for institutional-grade stock research
Last year, Stanford researchers built an AI and gave it the same public data any investor can access.
They let it adjust 3,300 mutual fund portfolios over 30 years.
It beat 93% of professional fund managers. By 600%.
Not because it found secret information. Because it processed public data without getting tired, emotional, or overconfident.
Norway’s $2.2 trillion sovereign wealth fund now uses Claude to screen every stock entering their portfolio. A ChatGPT-managed fund launched in 2023 is up 55%, beating the UK’s top funds by 18 points.
The tools exist. Most people just don’t know how to use them.
I spent months building prompts that produce the same analysis a $200K/year hedge fund analyst would deliver. Position sizing. Entry strategies. Risk matrices. DCF models. Bull vs bear debates with actual data.
The difference between useless AI output and institutional-grade research is entirely in how you ask.
________________________________________
What You get in this Resource
8 copy-paste prompts I run before every investment:
1. Institutional Investor Analysis — Why hedge funds would buy or avoid. Position sizing. Entry points. Crowding risk.
2. Bull vs Bear Debate — Two analysts arguing opposite sides with specific data. Each attacks the other’s best point.
3. Earnings Breakdown — Read quarterly reports in 10 minutes. Management tone. What they emphasized vs. avoided.
4. Risk Assessment — Business, financial, management, external risks. Probability ratings. Early warning signs.
5. Warren Buffett Analysis — Moat evaluation. Management quality. Owner earnings. Intrinsic value with margin of safety.
6. Earnings Call Analyzer — Paste the transcript, get the signal. What management said between the lines.
7. Insider Activity Analysis — What executives are doing with their own money. Cluster buying. Suspicious selling.
8. DCF Valuation — Three scenarios. Sensitivity tables. Probability-weighted fair value.
Plus: The exact workflow I use. How to chain prompts together. Mistakes that produce garbage. What changed in my process.
The math: A Bloomberg terminal costs $25K/year. These prompts produce comparable analysis for the cost of one dinner.

Next Round资本的创始人表示，最近几周，大约有六家机构投资者联系他的公司，寻求出售约6亿美元的OpenAI股票。在去年，这些股票通常几天内就会被抢购一空。但现在无人问津。目前市场对OpenAI的出价估值约为7650亿美元，较此前8500亿美元的估值折价约10%；对Anthropic的出价估值约为6000亿美元，较上一轮融资高出超过50%。
OpenAI表示，不认可也不参与这些交易。

第一次对Sam Altman和OpenAI有人性触角的感知是纽约客这篇人物文章，作为小说都很精彩，纽约客真的是一骑绝尘。