Jeremy Aube

Cloudflare is kinda annoying because they are so good it's hard to ignore and I kinda want to use JS again just to use their Workers

I know I'm late, but Astro is kinda nice

There is no toiler paper left Shit issue xitter-newx.xyz-mikolabs.workers.dev/typecraft_dev/status/2…

I understood only recently that when people say they have "20+ agents running at the same time" they really just mean that they have maybe 2-3 active and the rest are paused, waiting for them to get back to it.

One thing that is really annoying with ChatGPT is that it references website or companies and doesn't link to their website. Instead it opens an AI generated description. I know it's for retention but c’mon, it's so annoying.

Might just vibe-migrate everything to Rust just to save token budget by not having to upgrade NPM vulnerabilities every 5 minutes

The web is such an incredible platform. I can't count the number of times I though it would be hard to implement a feature and it ends up being an HTML attribute or 10 lines of javascript. Absolutely insane

GLM 5.3 is one hell of a model for an open-weight model. I've been using it almost exclusively for the past 10 days and it's been really great. It goes a bit too far at times (Will push commits to main and stuff if you don't block it). But nothing that can't be fixed with a quick Pi plugin

Some people actually enjoy using Windows?

imagine how different the current AI situation would be if it wasn't for competition

The older I get, the more I realize just how much of a nerd I am. I truly just like understanding computers for the sake of it

All the Si-Fi movies about the robots taking over the world are in these LLMs training data, just saying

My probably bad prediction for the future of AI is that it's going to be (and already is in many ways) better than humans at specific domains, but those domains will be considered niches compared to human intelligence

Learning in the age of LLMs is harder and easier than it was before. Information has never been easier to access, but you also have this magic button that does the job and you learn nothing from it. Balancing current productivity with future productivity is hard

This Ornith 36b MoE model is actually quite good. I am a local LLM believer now

The software factory pattern doesn't really work in my experience. AI is very good at condensing information but not very good at producing it

I asked Grok 4.6 to refactor some terraform code. 15 minutes later and it started building a game in python...

Everything in the AI world is a lot simpler to understand when you finally notice that it's just the same old patterns you've been using for years, but with an LLM somewhere in there and a new buzzword

My love of programming was actually a love of software architecture. Understanding boundaries, how they talked with each other, who owns the data, etc. This is why I am having so much fun with AI coding, this is what I get to do most of the time now. The AI implements the architecture I define.