1/ Initial reactions after some hours with this groundbreaking result proving the NP-hardness of poly-approx CVP/NCP: It is most likely correct, but more importantly, it is original, elegant, and beautiful! (Also: it is easy to improve, quantitatively.) openai.com/index/ten-ad...

Ten advances in mathematics and theoretical computer science

OpenAI shares new results on long-standing open problems in mathematics and theoretical computer science, including advances in geometry, cryptography, and complexity.

openai.com

The GapCVP result from OpenAI is interesting. Though I would at least like to see some peer review... a 5 MiB lean file is a lot to parse.

Yes there’s a solar eclipse and a lunar eclipse coming up this month… but are you ready for 🥧 “Not Quite As Good Pi Approximation Day” 🥧 on 25/8?!

Happy for this long-awaited feature in GitHub: stacking pull requests github.blog/changelog/20...

Stacked pull requests are now in public preview - GitHub Changelog

Stacked pull requests break large changes into small, reviewable pull requests. They’re an ordered series of pull requests that each represent focused layers of your change. With stacks, you can…

github.blog

can I have one for *everything*

Joseph Cox@josephcox.bsky.social · 7d ago

New: LinkedIn just introduced a "seems like AI slop" button. That's literally what the button says. The thought leaders are going to have a meltdown www.404media.co/linkedin-int...

I'm guessing this was caused by vibe coding.

Maggie Harrison Dupré@mharrisondupre.bsky.social · last wk.

NEW: A large trove of Claude-generated documents have been exposed to the web and indexed by Google. We reviewed what appear to be real medical records, the full names and phone numbers of school-aged children, and company documents marked for internal use only. futurism.com/artificial-i...

Another interesting homomorphic encryption update: Niobium has released open source tooling for their accelerator niobium.co/press/open-s... It's got a DSL, an agent skill, and tooling for working with their ecosystem (eventually including heir.dev !)

Open-Sourcing Niobium FHE Developer Tools

We believe encrypted computation should be accessible to any developer. We deployed and licensed the first part of our development stack on GitHub under Apache 2.0.

niobium.co

Everyone talks about Shannon's playfulness but from what I've read I feel like there's a lot more to the story. A greatness inspired by broad exposure to problems of practical importance, followed by an acute focus and synthesis of ideas, later squandered by fame and a lack of incentive or pressure.

Zach Weinersmith@zachweinersmith.bsky.social · 3w ago

Nerd thing: so there's this tech/math/nerd culture that's hard to describe, but where there's a kind of enlightened playfulness, which if you like you can trace very far back, and which perhaps hits a kind of high watermark with Claude Shannon, routes through late 60s/70s hacker culture, and so on.

Belfort released a nice homomorphic encryption demo today: CIFAR-10 inference in CKKS with GPU acceleration in 163ms. With round-trip latency, it's 200ms. Live web demo here: sofar.belfortlabs.cloud Some more details on their blog belfortlabs.com/blog/sofar

By a 6-3 vote, America is a big kid and does not need bedtime. But by a 5-4 vote, America is NOT TIREDDDDD *punches the Statue of Liberty*

I don't know if this is uniquely Portland, but on public transit I tend to find glitter on the seats a lot more often than, you know, other more unpleasant things.

not swords to plowshares but dashes to underscores

fixed your headline

Nick (derogatory) ✨@slothropsmap.bsky.social · last mo.

They’re calling it the least comprehensible achievement of all time.

I was surprised to find out yesterday that I have a poster at STOC's "Can AI do theory" workshop pritishkamath.github.io/ai-tcs-stoc-... for this paper: arxiv.org/abs/2605.14718 > Adapting AlphaEvolve to Optimize Fully Homomorphic Encryption on TPUs I won't be at the event, but hey that's fun.

Cliff Stoll (kooky scientist, Cold War era hacker, klein bottles) published a 1995 essay and then a book claiming the web was overhyped and apparently it became known for being so wrong. But overall I think he really nailed it!!!!!

OK all you ML experts I have a dumb question. Context: I am looking at MobileNetV3, which takes a 2D convolution and splits it into depth-wise and cross-channel convolutions with a hard-swish in between.

(Some) writers appreciate LLM writing assistance but readers hate it. Survey says… writethatblog.substack.com/p/dev-reacti... If readers suspect AI, they not only stop reading, but also downvote/avoid the author.

Avg LLM user: INJECT THE TECHDEBT INTO MY VEINS

Good blog post by Suresh Venkatasubramanian (@geomblog.bsky.social) on the latest installment of the Spectacle regarding Anthropic and the US government. He brings the popcorn, but the receipts and analysis, too. blog.geomblog.org/2026/06/the-...

I'm having fun with this 1001albumsgenerator.com things, but it's really teaching me how picky I am with music. I have listened to ~50 albums so far and there was only really one album I enjoyed that I didn't know in advance I was going to enjoy (KIWANUKA by Michael Kiwanuka),

1001 Albums Generator

Explore the history of music. One album a day. From the book 1001 Albums You Must Hear Before You Die.

1001albumsgenerator.com

And so it begins (kids summer vacation with patchy childcare).

english isn't for cowards. if you're not saying 'torpify,' that's a skill issue. there's no Académie Française telling you what to do. if you try to type 'pallific' and freak out when you see a little red underline, maybe you're not ready to roll with the big dogs.

So is the default assumption now that any ridiculous-sounding software outage or security incident is caused by AI?

tokenmaxxing as a form of corporate sabotage

The exploit: 1. Change your IP address to the target's geolocation and tell the support you've been hacked, asking them to send a reset to attacker email 2. No step 2. Meta's CISO immediately resigned. Gergely had inside contacts: the code was written by AI, reviewed by AI, while tokenmaxxing

My 7-year-old made her first Cayley graph and Cayley table today! And she LOVED it! At 9pm, she was too excited to go to sleep…”but I just want to do more algebra!” 🥰 #MathSky #VisualAlgebra

The Duality Technologies team published a comprehensive SoK on private LLM inference via FHE (with a focus on the CKKS scheme). It’s still inefficient, but FHE has come a long way. I am very hopeful that we'll get there someday. eprint.iacr.org/2026/935

SoK: Private LLM Inference using Approximate Homomorphic Encryption

Although recent surveys on privacy-enhancing technologies concluded that FHE cannot feasibly evaluate non-linear activation functions in modern ML architectures, 20 CKKS-based frameworks have since de...

eprint.iacr.org

Wow! Alfred Menezes just published this 182 page "A Gentle Introduction to Lattice-Based Cryptography" paper. I just skimmed through it, but it looks like an invaluable resource if you want to study lattices and how they're used in (PQ) Cryptography. eprint.iacr.org/2026/1098

A gentle introduction to lattice-based cryptography

We present the quantum-safe Kyber key encapsulation mechanism (ML-KEM) and the Dilithium signature scheme (ML-DSA). We also develop the mathematical background on lattices needed to understand why Kyb...

eprint.iacr.org

Funnily enough, I was TOLD that this is how chemists do clustering, and then when I looked into it actually they do clustering with far more principle than computer scientists and statisticians.

Mattan S. Ben-Shachar@mattansb.msbstats.info · 2mo ago

Psychologists doing clustering be like: #statsmeme

Small batch RAM, made in Portland.