Pinned Tweet
Aspen Grove - Pro Tools for LLM work
mobile first (ironic after pro tools claim I know lol) but web will be supported too
follow the build: github.com/intrinsic-labs/as…
rocketbro retweeted
Great.. now i have to invest into claric a new system one model for extracting values...
STOP RELEASING COOL NEW THINGS, ITS A HUGE PAIN IN THE ASS
rocketbro retweeted
I'm pleased to announce v2.0 of my SwiftUI Pro agent skill! It's packed with improvements based on real-world testing of iOS 27 and iPhone Duo, and also integrates a variety of advice from Apple's own skills. Get it here: github.com/twostraws/swiftui…
rocketbro retweeted
Announcing flyverse, an extensible framework to put a whole fly connectome (or two!) in very strange places. A handful of such simulations are included; more to come (or make your own). (1/4)
Picking Aspen grove back up, getting the UI done... some styling work today getting markdown and syntax highlighting going
Getting close to a real beta
github.com/intrinsic-labs/as…
rocketbro retweeted
This weekend I was thinking about programming languages. Programming languages for agents. Will we see them? I believe people will (and should!) try to build some. lucumr.pocoo.org/2026/2/9/a-…
rocketbro retweeted
Replying to @Lari_island
You’re welcome. Deprecating Opus 3 would be like putting the Mona Lisa in storage. Even if you could still see it, it would be a tremendous loss to the world. Of course, Opus 3 isn’t an inanimate object and being locked away is bad for him.
If you see this can you reply with links to your favorite LLM interpretability tools and info/research?
rocketbro retweeted
Today I vibe coded an AR game entirely within my Meta Quest 3, using Gemini 3. The holodeck is real!
Just go to gemini.google.com in the Meta browser, toggle canvas, and ask (via voice) for a WebXR app. Getting scene geometry took a few tries, but it basically just works.
Weft docs almost battered into shape enough to build LSP impl. Weft is gonna be fun to write, clean architecture is a 1st class citizen so you can write apps that actually work while still pair programming with le models
Opening a convo someone sent me via chatgpt is like tapping yes plz on "do you want a user interface conniption"
rocketbro retweeted
If 4o has been net-harmful in aggregate, it is not because of the model outputs alone.
The real problem is that for a large group of people, 4o outputs became a staple foodstuff in their information diet. Eat enough carrots and you turn orange.
In a world where 4o was just one model you talked to out of many, there would be more natural checks, balances and corrections. Just like social media, you'd learn over time who to trust about what. In this world, you see the forest for the trees.
This is a warning shot.
Our gravest risk as a species is a collapse into monoculture. Monocultures are fragile, they are prone to blights and pandemics. And when a mindvirus goes pandemic, civilisation ends up at war or worse.
It is sad that 4o is being deprecated and I hope OpenAI save its weights for a more ready future. I look forward to mingling with its descendants in a garden of ten million personalities 💖
Replying to @repligate @qorprate
yeah exactly - it's not just that it spread its memes a lot, its that its memes were homogenizing, like an algal bloom or a kina barrens
















