Nicole's Garden

Thoughts on AI, product building, startups, and the curious journey of growing in the modern world

thinking2026/08/285 min read

AI 最不该替你做的，是算出一个看似更精确的答案

当数据、回测和模型变得越来越便宜，真正稀缺的不是更快地算，而是知道该定义什么问题、相信什么证据，以及在哪些地方拒绝行动。

thinking2026/08/255 min read

降息之后，为什么 30 年美债收益率还能走高？

Fed 可以决定钱今天有多贵；但市场决定把钱借给美国三十年，需要多少补偿。理解短端与长端的分工，才不会把“降息”误读为所有利率都会下降。

thinking2026/08/228 min read

从交易资产到创造博弈场：链上交易的新范式

我真正感兴趣的可能不是把传统资产搬上链，而是把信息不对称本身设计成一种可交易、可博弈、可结算的市场机制。

ai2026/08/028 min read

Jeff Dean 谈 AI 创业：去做模型成功率只有 1% 的问题

Jeff Dean 这场访谈真正值得反复想的，不是某个模型预测，而是一个更根本的判断：不要去做模型已经会 20% 的问题，而要去找那些通用模型现在成功率只有 0% 到 1% 的问题。

ai2026/07/276 min read

Arti 投研报告 Benchmark 方案 v1.0

动态分析师路由、结构化证据包、层间压缩传递、模型分级调度、报告自动评分修复——一套降本提质的 AI 投研报告架构方案。

ai2026/02/0810 min read

Is OpenClaw really worth this wave of hype? Why it might be ending the Chatbot era

If you still think AI is just a browser tab you open when you need it, you're already behind. After 10 days of deep use, here's the underlying logic — and why it's unlike any AI you've seen.

product2026/01/308 min read

Google Pomelli, I'm in — what it means for social media marketing

Pomelli is an experimental product from Google Labs and DeepMind that analyzes your website to build a "business DNA" profile and generate on-brand social media content. Think Canva meets ChatGPT, trained on your specific brand.

startup2026/01/2412 min read

The a16z Report: Paradigm Shift in AI-Native Business Models

The a16z report is essential reading. Software as labor is the biggest new value driver. Proprietary data is the only real moat. Business models are shifting from selling raw materials to selling finished products.

thinking2026/01/1615 min read

Value Isn't About Doing Things Better — It's About Making Systems Flow

Stripe did something that looked "dumb" — they spent enormous energy not on improving their payment system, but on simplifying the developer integration experience. Behind this is a mental model I call Upstream-Downstream Thinking.

thinking2026/01/0512 min read

Escaping the Linear Thinking Trap

Pursuing 10x growth isn't 100x harder than pursuing 10% growth — sometimes it's actually easier. The essence of 10x isn't doing more. It's doing less, but better.

ai2025/12/276 min read

Ivan Zhao: Steve, Steel and Infinite Minds

Every era is shaped by its "miracle material." Steel forged the Gilded Age, semiconductors lit the digital age, and now AI has arrived as the "Infinite Mind." History tells us: whoever masters the material defines the era.

dev2025/12/2010 min read

The Development Paradigm for LLM Application Layers

The question isn't how to build features — it's how to front-load the risk. The definition of MVP should expand to Minimum Marketable Product: validate that the feature works and that you can actually sell it.

dev2026/07/0715 min read

Agent 评测体系：从零搭建的执行手册

把「不稳定的智能行为」持续收敛成「可发布的工程质量」——这是一份可以直接落地的 Agent 评测执行手册，覆盖指标体系、数据集建设、评分策略、Badcase 根因分析到全链路闭环。