keeran

Project mode in Claude desktop with Opus 5.5 is quite something tbh. It's helping me build a thing I started with Astra, I decided to try a spike of an idea using it, and erm.. haven't gone back. I have Codex all over my projects, but this vibe of just pushing the thing forward is slick.

Convincing looking phishing attempt or a legit @OpenAI account email? Not heard of @onelogin before but they look real.

Can someone tell me when the WoW multiverse situation has been resolved pls?

Not really sure what the point of an in-page GitHub copilot is if it doesn't have the ability to help with moving issues around in projects, sort out labelling etc. First time I'd ever tried it hoping it would be able to and it looks like it's a service that links to docs. Soon?

OK so who's working on a herdr frontend in a WoW addon?

reliving the early days by trying to work with qwen and a 64k context then moving to ornith with 262k.

Everyone is about to become an eval specialist.

This might well be the defining moment of the singularity. We are yet to fully understand our own intelligence at a physical level. Neuroscience is our best effort. This category of RSI feels like it could blow the doors off really fking quickly.

Bro just killed a solid quarter of goodwill and enthusiasm @poteto and co have been building in a single tweet.

This Linux CATASTROPHE comes with a 'if you have MPTCP enabled' kinda fear exhaust. I have no idea what that is but all of my servers have it enabled. This is fucking BAD.

My job is evenly split into making sure our agents are working on the right thing, and trying to remember to implement Teams alerts that let everyone else know what's going on.

Tonight I had to handle an employee at a favourite watering hole attempt to convince me that 60K people in the UK were arrested last year due to their social media activity. A few of us tried walking through how that number became prominent. There was some shared awareness of manipulation in the media or coopting of stats to suit a narrative. Yet when presented with whatever I could source as 'credible', I was met with a pre-programmed 'thats what they want you to think' kinda response. There is a clear lag in credible statistics because it takes time for these archaic organisations to do their thing. Fine. But it opens up a massive hole of credible uncertainty! 2023 stats for x say y There's no data for 2024 yet. Your argument is for 2025 and 2026 though... Tommy or Johnny whatever, with their wetherspoons massif, cheap suits and salt-of-the-earth banter can make up whatever they want and noone can challenge them! The gov.uk people are, imo, some of the best in the game. I don't know what it takes to enable them to provide more current data, but I hope it happens soon.

Does Elon pay Apple to have X show up top organically when you search for ‘Twitter’ in the app store or are they just doing the world a solid for free?

Have been experimenting with using unlocked open models to plan shady work and delegating implementation to frontier models in ways that won’t trigger flags. Side benefit is a CLI tool that spins up isolated dev environments bound to a directory with args for model etc, but all bound to a central herdr instance that lets me monitor / drive them individually. The unlocked model creates development packets that (so far) are innocuous and are directory-segregated. A higher level plan knows what is being built but only the local model knows about it. Within herdr I see it all, but each session is living in an isolated Docker environment entirely unaware. Will polish and put on GH over the weekend.

The OpenAI dudes really coming out with the "doubling your limits" cheerleading and not the "shit ok OK we won't half your limits, sorry!"