"And you better start swimmin' Or you'll sink like a stone For the times they are a-changin'" My GitHub contributions are up 379% over the last 6 months areyougoingexponential.rhys.dev/shaunchurch

Noticed I reposted something I didn't intend to repost. Just so you know.

Where are we tracking the progress of open weight AI models, local models, powerful consumer chips for running things, hardware requirements, decentralized compute networks? I went to OpenAI, which sounded like the right place, but was very disappointed.

I'm getting pretty lazy with context management again in this 1m context window, but I bet I'm still getting worse results.

The Claude Code in tmux flickering like a disco ball thing seems to be fixed! It still scrolls through the entire buffer top to bottom sometimes, but I'm starting to find that quite endearing tbh.

A slow rollout of a new "step-change" model, Mythos, is a great way to fuel the speculation and hype. "Trust me bro, this one makes no mistakes. So powerful, sorry you can't afford it yet"

The tweetdeck layout of X Pro was literally the only reason I paid for this shit. Not paying 4.5x more for a web view. Time to build something to scrape my lists, then.

Claude consistently comes up with new ways to surprise me.

"Yeah, welcome to the club pal"

Jacob Collier vibe coding an orchestra youtube.com/shorts/Wfm-VzQ…

TIL you can run virtualised Tahoe on Sequoia with tart.run and it works *really well*

It's prompts all the way down.

The story: January: Claude is amazing, the world has changed. Max out the plan. Go a bit insane. February: they nerfed it or I doubt my sanity. Claude sucks now, or am I holding it wrong? March: Cancel Claude sub and go back to GPT-5.4. April: The world has changed again but

OpenAI offering extra usage is just to get me hooked on /fast mode isn't it.

When GPT 5.4 reviews Opus 4.6 it says: Should try harder. Fix these things, and maybe don't do this. When Opus 4.6 reviews GPT 5.4 it says: This is over engineered, maybe don't try so hard.