incanter of tations 🔦🥽
@ArghZero
Followers
217
Following
4K
Media
66
Statuses
640
🌍 indie dev in taipei building tools to make the internet work for us, one app at a time 🌳💬 @treechatai 🐦🐻 @BirdBearApp
Taipei, Taiwan
Joined May 2020
sucks that essays are a dying art form. knowledge transfer hasn't slowed, if anything twitter+tiktok means more knowledge spreading overall. but we dont have a new medium for transmitting dense/deep knowledge yet. weve just optimized superficial knowledge transmission crazy hard.
these men were the great thinkers and writers of the 2010s. their essays massively influenced culture and thought and the arc of the world in very tangible ways. where are the new great essayists of the 2020s? who's out there building monumental public bodies of work?.
0
0
0
quick way to switch between regular anthropic claude code and claude code on s new claude code-ish plan (.
@steipete handy wrapper to launch claude in glm mode:. ```.#!/usr/bin/env bash.export ANTHROPIC_BASE_URL=export ANTHROPIC_MODEL="glm-4.5".export ANTHROPIC_AUTH_TOKEN=. exec "$@".```.put it on your path as glm and then you can do `glm claude` when you want glm.
0
0
0
👀 my anthropic days might be numbered. .
Announcing GLM Coding Plan for Claude Code!. After seeing the amazing adoption of GLM-4.5 over the past month, we're making it more accessible. Get started: Integration guide: What's new:.1/7th the price of original Claude Code
0
0
0
💯
@TheAhmadOsman already happening but with a twist. bootcamps are pivoting to 'ai-native development' - teaching system design, prompt engineering, and agent orchestration instead of just syntax. the new baseline is coming in with ai fluency from day one.
0
0
2
quite interesting, learning to "code" is soon just going to be about learning the environment, tools and best practices.
@Heliocene Actually mode higher level and necessary fundamentals. It won't even be about architecture as much as Software Engineering tools (e.g. source control, dotenv, basic git commands, etc).
1
0
2
since it's a fork of openai's codex it works with gpt-5 through your chatgpt plus plan automatically. pretty slick!.
github.com
Fast, effective, mind-blowing, coding CLI. Browser integration, multi-agents, theming, and reasoning control. Orchestrate agents from OpenAI, Claude, Gemini or any provider. - just-every/code
0
0
1
just-every/code is seriously underrated -- a fork of openai's codex cli, but with a ui that doesn't make you want to poke yoru eyes out and some crazy cool features on top:. - it can connect to your web browser and control it!!.- can call claude and gemini clis to do multi-model!.
1
2
2
cloudflared is amazing, if you're not using it and you're at all interested in self-hosting you probably should be. 5 minutes of setup and you get a rock solid external connection to a pi (or whatever) running in your house, with ssl/tls and all. basically ngrok++, but permanent.
1
0
1
for the uninitiated, this is what i'm using to make claude code talk to non-anthropic models. seems to be the best supported way to to do it atm.
github.com
Use Claude Code as the foundation for coding infrastructure, allowing you to decide how to interact with the model while enjoying updates from Anthropic. - musistudio/claude-code-router
0
0
0
that said, in my (brief) testing so far it looks like actual usage is adding up a whole lot faster than that table above would suggest.
1
0
0
some (not at all verified) llm math on how costs would work out:. theoretically $100/mo gets you about 152 hours of coding time per week, which is not quite as much as you'd get paying anthropic $100/mo, but is not too bad either!
1
0
0
just hooked up claude code to zai's GLM model using claude-code-router. pretty impressed, so far it Just Works™!. this could actually be an good alternative to anthropic's max plans now that they got nerfed today. gonna keep using it for the week and see how it compares.
GLM 4.5 w/ Claude Code is PHENOMENAL. and now that Anthropic's weekly limits are about to go in action. i think i am ready to cancel my subscription and fully rely on GLM 4.5 at home for my agents.
1
0
1
even better, it works with claude's --continue option, so you can basically split one claude session into two and get twice as much stuff done without losing your painstakingly collected context.
0
0
0
double tapping escape in the prompt box in claude code is ridiculously underrated. lets you go back up through your history to "fork" your convo, so you can re-use your existing context.
1
2
1
cool read on mitm-ing claude code to reverse engineer why it work so much better than other ai agents,. tldr: anthropic is constantly reminding claude what it’s doing and which tools to use.
medium.com
Notice how Claude Code just… gets coding work done!
0
0
1
anthropic vs openai is apple vs microsoft all over again. microsoft: fundamentally lacks "taste" but technically skilled.apple: all about aesthetics, tech is means to an end. openai: focuses on benchmarks and product features.anthropic: all about vibes, tech is means to an end.
1
0
2
anthropic's upcoming weekly claude code rate limits don't make a lot of sense. if you use a lot of opus buying two $100 plans is actually a much better deal than one $200 plan!. one $200/mo plan gets you 24-40 opus hours per week, but two $100/mo plans gets you 30-70. what?
0
0
1
streams are actually crazy underrated. they’re everywhere, half the world spends half their time scrolling them, but somehow we’ve all just accepted that the people in control of “the algorithm” are the ones who will control what we see. why can’t we control our own streams?.
0
0
0