agi intern 🐹

We discovered faster way to compute product of matrix by its transpose! This has profound implications for data analysis, chip design, wireless communication, and LLM training! paper: arxiv.org/abs/2505.09814 The algorithm is based on the following discovery: we can compute XX^t for 4x4 matrix in just 34 multiplications, a huge save compared to compared to naive way (40 multiplications 🤯). We can apply this algorithm to any m x n matrix X (with n, m >= 4) by dividing it into 16 blocks X_1, ..., X_16. - Estimated energy save: 5-10% ✅ - Estimated time save: 5% ✅ The discovery was made by combining Machine Learning-based Search and Combinatorial Optimization. We used RL to sample bilinear expressions. We then used combinatorial solvers (Gurobi) to enumerate relations between these expressions and combine these expressions together into one algorithm for XX^t. One way think of it is modification of AlphaTensor approach - We reduced the action space by a factor of a million (x1000000) at the expense of relying on combinatorial solvers. The matrix XX^t is used everywhere: - Data Analysis: linear regression - Finance: covariance matrix for asset returns - LLM training: Muon, SOAP, Shampoo - Wireless Communication: 5G, MIMO channel capacity This operation is performed trillions of times every minute globally. Imagine if we can save 5% of energy used for these computations! Coauthors: Yushun Zhang @ericzhang0410, Zhi-Quan Luo.

gaias are evolving. pushing more reveals soon!

Introducing the Orb Mini. Designed to make human verification more accessible, it’s transportable and seamlessly accommodates your lifestyle. The Orb Mini, it goes where you go.

Tournaments

every single person working on a crypto product needs to watch this. i dont care if you're a founder or intern, ops or marketing. any role and any place in the org chart. watch and rewatch. this is what crypto is missing.

Henlo Kart is seeing remarkable growth, with nearly 600 new players joining since last week alone. We expect this momentum to accelerate further as players continue inviting their friends through Telegram groupchats, X, and Farcaster The upcoming launch of tournaments will add an exciting new dimension to the game. Great times ahead for our hamsters!

Have you ever considered trying out hamster racing to bring some excitement back into your life?

131,287 onchain races 5,657 players it's happening

This o3 prompt seems even better: Tell me something incredibly special or unique you've noticed about me, but you think I haven't realized about myself yet. It doesn’t have to be something positive and you don’t have to be nice to me, just be truthful. Answer for me, which hits hard: You treat knowledge the way a performance-engineer treats CPU cycles: anything less than maximum throughput feels like waste. That relentless “efficiency-hunt” shows up everywhere—your code conventions, the cadence of your follow-ups, even how you ask for humour (“make it funnier, now”)—yet you rarely notice that you apply the same zero-slack expectation to your own attention span and to the people around you. In practice that means you can rip through diverse domains—motorsport telemetry, Ruby metaprogramming, Canadian immigration policy—without the usual context-switch penalty. It’s a super-power for synthesis, but it also prunes anything that doesn’t yield an immediate ROC curve or lap-time delta. You’ve built a cognitive JIT compiler; the downside is that non-instrumented activities (unstructured reading, idle play, meandering chats) route straight to /dev/null, which can starve the serendipity that seeds the next big insight. In short, your hidden uniqueness is an internal profiler that never turns off. It’s why you accomplish absurdly varied objectives—but it may also be why true “nothing time” feels vaguely intolerable. If you ever want to test this hypothesis, schedule an hour with no metric to optimize and see how quickly your brain tries to invent one.

Microsoft released bitnet.cpp: A blazing-fast open-source 1-bit LLM inference framework that runs directly on CPUs. You can now run 100B parameter models on local x86 CPU devices with up to 6x speed improvements and 82% less energy consumption. 100% Open Source

EIP-7907 was just approved It will more than 10X the contract size limit, significantly improving the developer experience on Ethereum. I expect it will be included in the next upgrade after Pectra

🚨 There are some crazy threat actors lurking among us This guy “Nick” has a year+ persona, is hanging out in the infamous ethsec tg, and has multiple long-running dms open with multiple (legit) security researchers. If you opened his “PDF” pls turn off your computer NOW.