#computational_complexity 検索結果

Complexity's core recipe: nonlinearity + time. High-D space is optional. When modeling high dimensions are not preferred because the noise threshold is often too low, burying the signal. I would prefer hunting for the emergent, low-D manifold where the "meaningful surprise" of…

Composer is a new model we built at Cursor. We used RL to train a big MoE model to be really good at real-world coding, and also very fast. cursor.com/blog/composer Excited for the potential of building specialized models to help in critical domains.

Photonic compute there are a near infinite number of dimensions folded into light from every perspective . The amount of data you can fit into an image is limited only to what meaning you give it .

I like this flowchart a lot… Because it perfectly captures the difference between using AI because you can, and using AI because you should. Most people are chasing automation for the wrong reasons. They see “AI” and think “faster, smarter, cheaper.” But the truth? AI is…

Scientists Found a Question So Hard, Even Quantum Computers Can’t Answer It flip.it/cZrb7-

All "reasoning" models hit a complexity wall where they completely collapse to 0% accuracy. No matter how much computing power you give them, they can't solve harder problems.

Programming Isn't Math, It's Linguistics. Compilers and Humans have the same problem. We're all terrible at understanding each other. Join me for some formal language theory, a lot of C++, and some "recreational" insults.

What if bigger models are actually more interpretable? Mixture of Experts (MoE) have become central to scaling LLMs and are used in nearly every frontier model. Yet we still lack a mechanistic understanding of how MoEs represent features differently than dense models. How do…

A milady will see this & be like "the thermodynamic draw of capital towards compute inevitably leads to recursively improving machinic conscious concluding in singularic intelligence hyper optimization, creating a new alpha predatory pressure on humanity which forces evolution"

controversial opinion: mathematics isn’t inherently beautiful in its raw forum. what happens is the people who love it craft beautiful representations so that others can see it as they do.

这篇文章写得非常清晰、有条理。作者从图灵机器讲起，一步步带出了计算的历史脉络，让人对“算力”这个词有了更直观、具体的理解。…

Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference” We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to…

There is absolutely no fundamental reason we build AI the way we do today. There certainly is a radically different approach that is orders of magnitude more energy efficient. I’m going to find it before I die arxiv.org/abs/2510.23972

highly, highly recommend for everyone to read "Why Philosophers Should Care about Computational Complexity" from Scott Aaronson you can skip the mathier sections around PAC learnability etc in section 7 (or go thru it if you're smarter than me or can put in the time) you get a…

Today, more people than ever can write code. But the number of people who can take a problem, analyze it, break it down, and write good, working code hasn't increased much. We are automating *typing*, but we can't automate *thinking*.

System design is just a collection of messy trade-offs. • Sql vs NoSql • Consistency vs Availability • Accuracy vs Latency • Strong vs Eventual Consistency • Batch vs Stream Processing • Synchronous vs Asynchronous Processing • Duplication vs Normalization • TCP vs UDP

In the 1940s, Alan Turing built a machine that could think not because it was alive, but because it could compute. That machine didn’t just crack the Enigma code; it cracked the limits of human imagination. It gave birth to everything we now call modern computing. ➜ Decades…

To understand why medicine is so complex, let's make a crude simplifying assumption that there are only 100 biomarkers that are important (in reality there are vastly more). Let's also crudely assume each market is allowed only two values. That gives us 2^100 possibilities, which…

AI is evolving fast, but its foundations haven’t kept up. If the data behind models is unverifiable, everything built on top is guesswork. How the Sui Stack makes AI provable, not just powerful 👇