4261
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 6,334 posts.
4262
From 300KB to 69KB per Token: How LLM Architectures Solve the KV Cache Problem
How the KV cache gives every AI conversation a physical weight in silicon, and what happens when the memory runs out.
4263
There is no hard takeoff
Back in 2014, Elon Musk referred to AI as summoning the demon. And it wasn’t hard to see that view. Soon, Go agents would beat top humans learning from self play. By the end of 2017, the same algorithm mastered Chess and Shogi. By 2020, it didn’t even need tons of calls to the simulator, and could play Atari too.
4264
4265
4266
4267
The real data wall is billions of years of evolution
Careful with those human analogies
4268
Hetzner is working on LLM Inference
Hetzner has launched an experimental LLM inference API. I tested its Qwen model—and have a few guesses about where the product could go next.
4269
New serious vulnerabilities spiked around release of Claude Mythos Preview
Notable organizations disclosed ~1,300 high- and critical-severity CVEs in June 2026, roughly 3.5× the pre-Mythos monthly record. Epoch AI's breakdown of vulnerability disclosures following Anthropic's Project Glasswing.
4270
How is Groq raising more money?
Somehow, Palpatine returned.
4271
Xanadu was waiting for agents
From the Zed Blog: Ted Nelson specified the docuverse sixty years ago. The technology finally arrived, and so did its native users.
4272
Logistic regression from scratch
Deep dives into Cryptography and Privacy.
4273
Things I've learned in my 7 years implementing AI
Even though the impacts of LLMs have never been seen before, they feel familiar to earlier assumptions.
4274
Gemini API File Search is now multimodal
Updates to the Gemini API File Search tool makes building efficient, multimodal file retrieval systems easier for developers.
4275
4276
4277
4278
Isolating Football Player Movement by Eliminating Camera Motion
Note: This post departs from the general format used on this blog. That's because the post is not about a specific analysis I've done bu...
4279
Deep Learning: Our Miraculous Year 1990-1991
Many of the basic ideas behind the Deep Learning Revolution were published three decades ago within fewer than 12 months in our "Annus Mirabilis" 1990-91
4280
Cuda.jl v3.3: union types, debug info, graph APIs
There have been several releases of CUDA.jl in the past couple of months, with many bugfixes and many exciting new features to improve GPU programming in Julia: CuArray now supports isbits Unions, CUDA.jl can emit debug info for use with NVIDIA tools, and changes to the compiler make it even easier to use the latest…
4281
Biohub releases a world model of protein biology
Open AI models used for a wide range of scientific applications, including accelerating the design of therapeutic molecules.
4282
How to scale LLMs better with an alternative to transformers
Monarch Mixer: Revisiting BERT, Without Attention or MLPs
4283
Using LLMs to Generate Fuzzers
LLMs seem surprisingly good at many things. So much so that not a week goes by without someone coming up with yet another use-case for this technology, often to solve tasks quickly that traditionally took a non-trivial amount of human work to complete. Today’s example was Brendan Dolan-Gavitt’s remarkable experiment…
4284
Reverse Engineering Google Colab
Take a peek into the internals of Google Colab, and discover how we can bend the rules of Colab a little in order to adapt it to our MLOps workflow
4285
Voxtral – Frontier open source speech understanding models
The most powerful AI platform for enterprises. Customize, fine-tune, and deploy AI assistants, autonomous agents, and multimodal AI with open models.
4286
4287
The Three Cultures of Machine Learning
Everyone should read Leo Breiman's 2001 article, Statistical Modeling: The Two Cultures . (Summary: Traditional statisticians start with a distribution. They try to identify the parameters of the distribution from data that were actually generated from it. Applied statisticians start with data. They have no idea where…
4288
Zi2zi: Master Chinese Calligraphy with Conditional Adversarial Networks
zi2zi is the follow-up work for my last project , once again tackling the same problem of style transfer between Chinese fonts.
4289
4290
Building a real-time AI tutor for 5-year-olds
We set out to build the first AI tutor to teach math and reading to kids ages 4-9. For AI to actually teach a five-year-old, pedagogy must be baked into the engineering. A child can't wait for a slow reply, can't read a chat interface, and can't unhear anything a model gets wrong. We wanted to share some of the…