5041
4000x Speedup in Reinforcement Learning with Jax
Web publication for "PureJaxRL"
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 6,335 posts.
5042
Voynich Manuscript: word vectors and t-SNE visualization of some patterns
Update 17/01: reddit discussion thread. Update 19/01: hacker news thread. The codex The Voynich Manuscript is a hand-written codex written in an unknown system and carbon-dated to the early 15th century (1404–1438). Although the manuscript has been studied by some famous cryptographers of the World War I and II…
5043
5044
Phi-4 Reasoning Models
Microsoft continues to add to the conversation by unveiling its newest models, Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning. Learn more.
5045
Julia GPU
Lambdaclass's blog about distributed systems, machine learning, compilers, operating systems, security and cryptography.
5046
Siri, Privacy, and Trust
We should expect Apple to lead the industry on this front, but in fact, they’re far behind.
5047
5048
Inverting PhotoDNA
Ribosome inverts Microsoft PhotoDNA hashes using machine learning.
5049
Pixtral 12B
Pixtral 12B is deprecated. Discover Mistral AI’s latest vision models and capabilities in our updated documentation.
5050
Four pitfalls of hill climbing
An animated look at local maxima, novelty effects, differentiation, emergent maxima.
5051
5052
Generating Recommendations at Amazon Scale with Apache Spark and Amazon DSSTNE
In this post, I discuss an alternate solution; namely, running separate CPU and GPU clusters, and driving the end-to-end modeling process from Apache Spark.
5053
Singular Value Decomposition Part 2: Theorem, Proof, Algorithm
I’m just going to jump right into the definitions and rigor, so if you haven’t read the previous post motivating the singular value decomposition, go back and do that first. This post will be theorem, proof, algorithm, data. The data set we test on is a thousand-story CNN news data set. All of the data, code, and…
5054
Tversky Neural Networks
Psychologically Plausible Deep Learning with Differentiable Tversky Similarity
5055
DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression
A deep dive into the DeepSeek-V4.1 Flash technical report: CED, CSA2, HSI, Single-Pass mHC, Engram and FP4 KV Cache — how the KV cache was compressed to just 890 bytes per token.
5056
5057
5058
"ChatGPT said this" Is Lazy
When you paste ChatGPT’s response instead of your own feedback, you’re not being helpful. You’re being lazy and creating more work for everyone.
5059
What Happens When the Cost of Intelligence Drops 100x
A given level of LLM intelligence costs about a thirtieth of what it did a year ago, and the rate of decline is accelerating. Using Artificial Analysis's…
5060
5061
If writing is thinking then what happens if AI is doing the writing and reading?
Then what happens if AI is doing the writing and reading?
5062
Hostile AI: You're soaking in it
I was in a Facebook discussion about “Friendly Artificial Intelligence” — this is a buzzword from the Singularity Institute people. They be...
5063
We built an AI-powered Magic the Gathering card generator
Cohere's platform was used to create an AI that would generate playable Magic The Gathering cards based only on a given name of each card.
5064
Stop using natural language interfaces
Natural language is a wonderful interface, but just because we suddenly can doesn't mean we always should.
5065
I turned Markdown into a protocol for generative UI
An AI assistant that builds reactive UIs mid-conversation — forms, streaming data, callbacks — using markdown code fences and a mount() UI primitive.
5066
Exploring the mysteries of Go with AlphaGo and China’s top players
Just over a year ago, we saw a major milestone in the field of artificial intelligence: DeepMind’s AlphaGo took on and defeated one of the world’s top Go players, the legendary Lee Sedol. Even then, we had no idea how this moment would affect the 3,000 year old game of Go and the growing global community of devotees…
5067
How to Optimize a CUDA Matmul Kernel for CuBLAS-Like Performance: A Worklog
In this post, I’ll iteratively optimize an implementation of matrix multiplication written in CUDA.My goal is not to build a cuBLAS replacement, but to deepl...
5068
AI predicts certain esophageal and stomach cancers three years before diagnosis
AI can predict certain forms of esophageal and stomach cancer Michigan Medicine study says.
5069
Deep Agents
Build deep agents that plan and execute complex tasks over longer time horizons. Learn the architecture behind Deep Research and Claude Code.
5070
Facial hair tracking for high fidelity performance capture
Facial hair is a largely overlooked topic in facial performance capture. Most production pipelines in the entertainment industry do not have a way to automatically capture facial hair or track the skin underneath it. Thus, actors are asked to shave clean before face capture, which is very often undesirable. In this…