Also looking forward to knowledge-sharing between software and hardware formal verification

Truly exciting. Waiting to see more examples of formal bug-hunting

I created the image using my clawbot (TinyClaw) running on my Mac Mini

I invented a new term today - ‘Claw Computing’.

I re-launched a site I had started with a different objective more than a year ago. This time the main focus is around ways to use tools like Claude Code better. blog.codepromptfu.com/posts/codeprom…

@hboon.com here's that blog post I was telling you I'd write some day

Captured my experience and experiments with building an AI coding agent on my latest blog post debamitro.github.io/blog/coding-...

GLM 4.7 is great, and GLM 5 is blazing hot

1. The same people might be showing up with more humility and curiosity in other fora, possibly closed communities. Speaking from my personal experience

2. Only the loud voices get heard in any community. It is possible that’s why only the ‘leaders’ show up in our feeds

Is it easy enough to put React in a SwiftUI web view?

I think so, it’s working fine for coding tasks, it keeps going on through errors until it succeeds

Are you using Tauri?

Use GLM as the LLM from whatever CLI agent you use

Request for help: do I know anyone who ships electron apps? I'm stuck on a packaging/testing issue 😅

I’m not the CSS expert! I’m not even forming one, I’ll wait for the consensus.

Last year I dabbled a bit with the paged media extension, couldn’t make much progress. If this pretext thing is better I’ll use it

@nearestnabors.com what’s your take on this stuff which is rocking X now chenglou.me/pretext/edit...

Right. And using a descriptive system prompt might prevent that from happening, even if it is not always needed.

I am seeing the differences again. I think the system prompt _does_ play a role. I think sometimes the problem goes away because of some kinda cache at the LLM provider's end

Schedule one?

I still find it hard to believe that my computer is talking to me over Telegram messages. This should have been invented years ago!

I am learning more React from seeing what my AI agent does than I ever did from any human!

With the power of AI coding tools, I feel like Archimedes, who famously said 'Give me a lever long enough and a fulcrum on which to place it, and I shall move the world'. AI is the lever and all I need is the problem, i.e. the fulcrum

Then a VPS might be the best. Pieter Levels claims he keeps Claude Code running on the VPS and lets it handle traffic spikes or app issues. He apparently doesn’t even use github - the code lives on the VPS and is modified by Claude

Vercel is pretty smooth for full-stack nextjs. Their free tier is generous, so far. For static websites I like my $4/month VPS - it has been handing 5 websites

Last night I had that recursive moment when a coding agent I vibe coded fixed it’s own source code!

Actually I found a bug in my coding agent. Maybe my observation was not correct - the system prompt doesn't matter so much any more for the frontier LLMs

“Build like everyone’s buying” is the new “Dance like nobody’s watching”

Omg omg Meta acquired Moltbook!! I have to post this everywhere

Remotion is so much fun - specially when Claude Code controls it

Guru - Pristine Advice for Navigating Life's Demands Pristine advice from a guru to help you navigate life's demands. Come here if you seek happiness, money, peace or revenge

Created my first telegram bot. It’s a whacky advisor. Two years ago this was the first LLM-powered application I ever made and I launched it as a website. Today it is obvious that it should have been a bot all along.

Here’s the website which has the link to the bot lookingforin.life