1/ Today, we’re excited to introduce Attie, currently as an invite-only closed beta. Attie is the first agentic social app on atproto. It’s something completely new—an experiment in making building on the protocol more accessible.

👀👀👀

Maia Chess, the website and service based around Maia, the chess model I use for blunder.clinic, just got a big update.

Strong start and a very weak finish. blunder.clinic #19 • 1200 🟩🟩🟨🟨🟥🟥 6/6 ⭐ 💪 blunder.clinic #19 • 1500 🟩🟩⬜⬜⬜🟥 3/6 💪 blunder.clinic #19 • 1800 ⬜🟩⬜🟨⬜⬜ 2/6 💪 #chess #puzzle

Arrival

WE ARE SO BACK

it's over

Is everyone at @atmosphereconf.org already familiar with the Pac-Man Rule for Conferences? 😅 www.ericholscher.com/blog/2017/au...

Always an honour to get featured on @simonwillison.net’s site 🥲🫶 simonwillison.net/2026/Mar/26/...

wow

blunder.clinic #18 • 1200 🟩🟩🟨🟨⬜🟥 5/6 💪 blunder.clinic #18 • 1500 🟩🟩🟨🟨🟥⬜ 5/6 💪 blunder.clinic #18 • 1800 🟩🟩🟨🟨🟥🟥 6/6 ⭐ 💪 Finally got a 6/6 WHEW #chess #puzzle

blunder.clinic #17 • 1200 🟩🟩🟨⬜🟥🟥 5/6 💪 blunder.clinic #17 • 1500 🟩🟩🟨🟨⬜🟥 5/6 💪 blunder.clinic #17 • 1800 🟩🟩🟨🟨⬜⬜ 4/6 💪 Updated the #puzzle curation a little and fixed a duplicate puzzles bug. #chess

This is a crazy good post, even by the @samwho.dev standard.

I spent 2 months learning about quantization and am extremely proud of the post I've written about it. I think these are some of the nicest visuals I've ever made, and I love how this compression technique invented in 1898 is being used on the bleeding edge in 2026. ngrok.com/blog/quantiz...

our new Olmo Hybrid model combines attention with linear RNN layers 🍣training efficiency is crazy good. the model reaches same MMLU score as Olmo 3 in 50% of the tokens. also see this in many other tasks as always: weights, data, ckpts, training code, etc. all fully open

blunder.clinic #15 • 1200 🟩🟩🟨🟨⬜⬜ 4/6 💪 blunder.clinic #15 • 1500 ⬜🟩🟨🟨🟥⬜ 4/6 💪 blunder.clinic #15 • 1800 🟩🟩🟨🟨⬜🟥 5/6 💪 #chess

NeurIPS will hold a satellite location in Atlanta, Georgia, USA for those who cannot travel to Australia.

blunder.clinic #14 • 1200 🟩🟩🟨⬜🟥🟥 5/6 💪 blunder.clinic #14 • 1500 🟩🟩🟨⬜🟥⬜ 4/6 💪 blunder.clinic #14 • 1800 🟩🟩🟨🟨⬜⬜ 4/6 💪 One of these days I'll get the 6/6. #chess

Me and my rental car.

Now I'm seeing it everywhere and I can't tell if it's Baader-Meinhof or if it just actually broke containment and is entering the common lexicon.

After all these years, I am still basically incapable of fluidly reading romanized Korean.

Yo

Me and my rental car.

blunder.clinic #12 • 1200 🟩🟩🟨🟨🟥⬜ 5/6 💪 blunder.clinic #12 • 1500 🟩🟩🟨⬜🟥🟥 5/6 💪 blunder.clinic #12 • 1800 🟩⬜🟨🟨🟥⬜ 4/6 💪 So close on 1200/1500. #chess

Naver AI has put the word "World" into "World Models", at least on a metropolis scale: A world model for Seoul using Naver Maps (the Seoul capital area is 25.6 million people, btw.). seoul-world-model.github.io arxiv.org/abs/2603.15583 @jnhwkim.bsky.social

blunder.clinic #10 • 1200 🟩🟩🟨🟨⬜🟥 5/6 💪 blunder.clinic #10 • 1500 🟩🟩🟨🟨🟥⬜ 5/6 💪 blunder.clinic #10 • 1800 🟩🟩🟨⬜⬜⬜ 3/6 💪 Pushing some updates to the ux soon.

blunder.clinic #9 • 1200 🟩⬜🟨🟨🟥🟥 5/6 💪 blunder.clinic #9 • 1500 🟩🟩🟨⬜⬜🟥 4/6 💪 blunder.clinic #9 • 1800 🟩🟩🟨⬜🟥🟥 5/6 💪 Better this time. #chess

I'm actually really happy this happened because it is a very nice example of my "everything in NLP is machine translation" mantra.

Kagi Translate for leetcode

New blog post: A Decade of Slug This talks about the evolution of the Slug font rendering algorithm, and it includes an exciting announcement: The patent has been dedicated to the public domain. terathon.com/blog/decade-...

The first conference I ever went to (and it was in Canada!). CIAA holds a special place in my heart.

Please submit your research papers and join us at the 30th International Conference on Implementation and Application of Automata (CIAA 2026), to be held in Kingston, Canada from August 5 to 8, 2026! Details: research.cs.queensu.ca/ciaa2026/ #tcssky #mathsky

📢I'm organizing a BoF session at #EACL2026 called Tokenization & Beyond, aiming to gather researchers exploring tokenization and alternatives such as byte-level and pixel-based approaches. Sign up using the form if you're interested! #NLProc @eaclmeeting.bsky.social

blunder.clinic #8 • 1200 🟩🟩🟨⬜⬜⬜ 3/6 💪 blunder.clinic #8 • 1500 🟩🟩🟨⬜🟥⬜ 4/6 💪 blunder.clinic #8 • 1800 🟩🟩⬜🟨🟥⬜ 4/6 💪 Rough 1200 #chess

I spotted a suspicious looking G-Wagon in Osaka, but something looked familiar. #ジムニー

The Enclose Horse Tapestries. 1495–1505, Wool warp with wool, silk, silver, and gilt wefts

🚨New preprint! LLM teams are being deployed at scale, yet we lack the tools to predict when they’ll succeed, fail, or how to design them. Distributed computing faced the exact same questions and figured out how to answer them. We show those insights apply directly to LLMs 🧵👇

For the pinned post 🐴 Check out our simple puzzle game! Try to enclose the horse in the largest possible area with limited walls. To support the game, give this a repost!

blunder.clinic #6 • 1200 🟩🟩🟨🟨🟥🟥 6/6 ⭐ 💪 blunder.clinic #6 • 1500 🟩🟩⬜⬜🟥🟥 4/6 💪 blunder.clinic #6 • 1800 🟩⬜🟨🟨🟥🟥 5/6 💪 #chess

meet interesting people

blunder.clinic #5 • 1200 ⬜🟩🟨🟨🟥🟥 5/6 💪 blunder.clinic #5 • 1500 🟩🟩🟨🟨⬜⬜ 4/6 💪 blunder.clinic #5 • 1800 ⬜🟩🟨🟨⬜🟥 4/6 💪 Oof

Shared this yesterday but have to share it again. Just a phenomenal idea from Marco, and super well executed. And what a domain 🤌

blunder.clinic #4 • 1200 🟩🟩🟨🟨⬜🟥 5/6 blunder.clinic #4 • 1500 🟩⬜🟨🟨🟥⬜ 4/6 blunder.clinic #4 • 1800 🟩🟩🟨🟨⬜🟥 5/6 Close, but no cigar

Today I launched blunder.clinic, a realistic daily #chess puzzle app! Short 🧵

@mcognetta.bsky.social this is really good. I do puzzles every day, and this is was so much more engaging, and, fun!

Machine translation is tough to evaluate, partly because most of what you throw at is too easy. That doesn't at all mean that translation is solved; we're just not doing a good job finding interesting inputs.

🧵New paper: "Lost in Backpropagation: The LM Head is a Gradient Bottleneck" The output layer of LLMs destroys 95-99% of your training signal during backpropagation, and this significantly slows down pretraining 👇

blunder.clinic #3 • 1200 🟩🟩🟨🟨🟥⬜ 5/6