Testing an idea that I've had for a while, but I think is even better suited for atproto Internet, meet plts.at

Plates · Your menu, everywhere

Publish your restaurant's menu once, as structured data you own. Plates gives it a beautiful page, and the open network carries it everywhere else.

plts.at

Alright, the Codex Micro just arrived! Now lets see how I can use this not with Codex but with my own harness, and hopefully also forcefully add a microphone to this device :D

It's all downhill from here, hold on to your horses

When our corporate overlords play opposite-day for a conference, what a fucking joke

It seems like #OpenAI is planning to encrypt all the prompts that #codex uses on disk, we can already not see whatever the main agent sends subagents as it's encrypted, and they're sending ciphertexts to the endpoints for inference?! What in the world?! Why would you do this? We need introspection!

Microsoft displaying the prowess of vibe-coding yet again, is there seriously no QA left at MS?

If you are looking for Jerry's Map original panels go to ebay.com! There are a few being sold at auction right now! Thanks for the massive response to People Make Games!

It really is! Just achieved ~400 tok/s in single prompt text-generation! Text-only still, but still :) #DiffusionGemma feels like it has a really good foundation for local AI. Loading a 50GB model into VRAM + 1024 tokens inference in under 8 seconds in total, I think this changes everything :)

embedding-shapes@emsh.cat · 2mo ago

DiffusionGemma is something else, getting +200 tok/s in my Rust implementation without even having a look at the performance, and the model is relatively clever for its size. GPU utilization is so much better than pretty much every other local model available.

Monday, June 15, 2026: Jerry's Map Section N.IV jerrysmap.com

Incredible map by @jerrysmap.bsky.social beautifully presented by @itsquinns.bsky.social and @chrisbratt.bsky.social One imaginary map created over 40 years, as a "analogue generative art machine" essentially. Really interesting, inspiring and made be tear-eyed at the end, what a magical project.

DiffusionGemma is something else, getting +200 tok/s in my Rust implementation without even having a look at the performance, and the model is relatively clever for its size. GPU utilization is so much better than pretty much every other local model available.

Man, all GPT models ran via the ChatGPT subscription feels a lot dumber for the last 3-5 days, feels like they're about to release a new model

Man this sucks so bad...

Using #Firefox and #1Password and suddenly see an emoji picker where you should see your usernames? Seems Firefox version 150 added a new shortcut that overrides 1Password's shortcut, here's how to disable it so you can use 1Password properly again: emsh.cat/en/how-to-di...

📣 ELLIS | ELLIOT Summer School on Autonomous Driving 🚗 Super happy to announce the hottest event this summer. Apply now! 📅 September 15-18, 2026 📍Barcelona, Spain 🌐 pages.cvc.uab.es/ssadriving/ @ellis.eu, ELLIS-Barcelona, CVC-Barcelona, @valeoai.bsky.social ai.bsky.social, @elliot-eu.bsky.social

Got myself a Mac Mini so I could do CI builds of macOS apps at home. First time I left it and came back later, monitor received *something* but screen was black. Bit of troubleshooting later, turns out I hit my first macOS bug, I have barely used it! I have a feeling I'm in for a lot of fun...

I wrote a short post-mortem on the Bluesky outage we had on Monday. First time using pckt, it was a really nice experience tbh! pckt.blog/b/jcalabro/a...

April 2026 Outage Post-Mortem - Jim's Pckt

Hey all! I'm Jim, and I do system-y things at Bluesky. I'm here to give you some details about what happened on Monday of this week that caused Bluesky to go...

pckt.blog

I'm gonna hazard a guess and say this won't ever actually finish

Big news! Igalia (@igalia.com) is partnering with @eurosky.social to develop the next generation of the open social web. Let's stop waiting for others to fix our problems and start fixing them ourselves. We can just do things. More: www.eurosky.tech/s/Eurosky-Ig...

Sherif EA@sherif.eurosky.social · 5mo ago

I am very excited to announce our partnership @eurosky.social with @igalia.com! The partnership will accelerate the development of independent, European-operated infrastructure for the open social web, enabling developers, organizations, & communities to build new applications. cc @xanlopez.xyz

I'm finally ready to show off the thing I've been working on: atBB (pre-alpha)! It's a forum platform built on top of atproto. I wrote up some thoughts about why I am interested in building forums: pckt.blog/b/mal-s-musi... 1/2

What the World Needs Now is Forums Sweet Forums - Mal’s Musings

and other thoughts from a forums purveyorI was seven years old when I got my first video game console. It was Christmas, and I actually got two! A GameBoy Po...

pckt.blog

#reddit mobile app seemingly got rid of "All" and you cannot even manually go to r/all... I guess it was about when, not if, and now I've finally uninstalled reddit and will stop using it, as they clearly want less users.

Is there no compile-to-Rust languages out there? Kind of surprising, why so little of them?

How would people feel about an open source/open data alternative to myheritage/ancestry and similar family tree services? So annoying they're inaccessible unless you register, shouldn't knowledge about past people be open?

So, played around openclaw and holy shit, its an excellent demonstration of everything wrong with "modern software engineering". Errors galore, setup broke multiple times during onboarding, technical debt all over the place, zero concerns with quality, connection breaks constantly, and so much more

What #investigative #journalists are currently digging through all the new Epstein files?

Tried #Kimi with Kimi Code because why not? And yeah, results aren't great. The goal? Make the build time faster

Today I'm excited to announce @sequoia.pub: a CLI for self-hosted blogs to publish @standard.site lexicons to the ATmosphere sequoia.pub/blog/introdu...

Introducing Sequoia: Publishing for the Open Web

sequoia.pub

Agents perform 50% worse when working in a team than working alone. The setup is simple: two agents, two tasks, two VMs, and one chat channel. They then evaluated whether the merged solution from both agents passes the requirements of both tasks.

Sounds like they need ATProto and ActivityPub

Open Future@openfuture.bsky.social · 6mo ago

European social media alternatives need more than EU hosting. They require decentralized architecture, robust safety mechanisms, public funding, and business models beyond surveillance. Read more in @aditya.myatproto.social latest: openfuture.eu/blog/what-is...

> running LLMs locally offer unparalleled room for experimentation. People are really sleeping on this, completely different experience when you can build software and just say "Fuck it, lets fire up 5 agents, then compare their work output against each other"

fry69@fry69.dev · 6mo ago

Mac Studio shine in (ab)using the vast unified memory for running LLMs locally (don't expect to run Opus 4.5 class at home though). It might still be more efficient/cheaper to rely on LLMs via APIs, but running LLMs locally offer unparalleled room for experimentation.