惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

量子位
博客园 - 三生石上(FineUI控件)
D
DataBreaches.Net
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
GbyAI
GbyAI
P
Proofpoint News Feed
Microsoft Security Blog
Microsoft Security Blog
月光博客
月光博客
I
InfoQ
V
Visual Studio Blog
罗磊的独立博客
Engineering at Meta
Engineering at Meta
Vercel News
Vercel News
Jina AI
Jina AI
L
LangChain Blog
Apple Machine Learning Research
Apple Machine Learning Research
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
B
Blog
The Cloudflare Blog
小众软件
小众软件
雷峰网
雷峰网
V
V2EX
人人都是产品经理
人人都是产品经理
Stack Overflow Blog
Stack Overflow Blog

Will Angel's Blog

Superhuman AI coding still makes software worse I did not say that, you did The AI Tarpit: Why You Can't Stop Reading Your Code Anthropic Fable The Stochastically K-Shaped Engineering Job Market Claude.AI Pro Plan quotas too small for deep research Apple Silicon costs more than OpenRouter Jankmarking: Janky Benchmarking Offline Agentic Coding: OpenCode Offline Agentic Coding Washington DC on track for most volatile temperature year since 1959
I cancelled my personal Claude subscription
2026-09-18 · via Will Angel's Blog

Tags: AI, Coding

Like this guy, Claude won't shut up.
Claude won't shut up

I cancelled my Anthropic subscription yesterday even though Claude Code is awesome.

Except it sucks at writing and communicating, and has been getting worse since around April. See the load bearing vocabulary for some quantitative examples.

Specifically, the vocabulary and linguistic structure that Claude drifts towards is mentally taxing to read and that makes Claude less helpful per unit of energy and time spent using Claude. Claude has gotten worse in this regard, and other options are almost as good and can be less verbose. All LLM writing is mediocre at best, but some of it is at least boring and straightforward. I do not want to read claudish.

Catch me using Kilo and open models for the near future.