661
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 3,091 posts.
662
Large-Scale Online Deanonymization with LLMs
We measure the capabilities of LLMs to deanonymize users online.
663
Will Smith's concert crowds are real, but AI is blurring the lines
Will Smith is being accused of generating fake fans with AI, but it's complicated: the crowds are real, but the videos were manipulated by Smith's team and YouTube itself.
664
Tokens are getting more expensive
"language models will get cheaper by 10x" will not save ai subscriptions from the short squeeze
665
AI slows down open source developers. Peter Naur can teach us why
Metr recently published a paper about the impact AI tools have on open-source developer productivity[^1]. They show that when open source developers working in codebases that they are deeply familiar with use AI tools to complete a task, then they take longer to complete that task compared to other tasks where they…
666
667
78% MNIST accuracy using GZIP in under 10 lines of code
Using GZIP compression and the k-Nearest Neighbors algorithm, we explore an innovative approach to classifying the MNIST dataset with about 78% accuracy
668
Image Kernels explained visually
An image kernel is a small matrix used to apply effects like the ones you might find in Photoshop or Gimp, such as blurring, sharpening, outlining or embossing. They're also used in machine learning for 'feature extraction', a technique for determining the most important portions of an image. In this context the…
669
GPT-5 Thinking in ChatGPT (a.k.a. Research Goblin) is good at search
“Don’t use chatbots as search engines” was great advice for several years... until it wasn’t. I wrote about how good OpenAI’s o3 was at using its Bing-backed search tool back …
670
I made a transformer to predict a simple sequence manually
To better understand how transformers work, I hand-assigned all the weights to predict a simple sequence.
671
Representation Engineering: Mistral-7B on Acid
Playing around with the Representation Engineering paper, I made some interesting control vectors, and a Python package to make your own.
672
673
I asked Meta’s Muse for its filesystem and it sent me 6.8GB
I asked Muse to archive the files it could see and send them to my Google Drive. It did.
674
Machine Learning in a Year
From being a total ml noob to start using it at work
675
676
How to effectively write quality code with AI
AI is rarely optional anymore, but how can you still be proud of your craft? Discover the workflow to effectively write high-quality, robust code using AI tools.
677
Making games in Go: 3 months without LLMs vs. 3 days with LLMs
After 15 years as a software engineer, I realized I had never actually built and published a game.
678
Prompt engineering vs. blind prompting
"Prompt Engineering" emerged from the growth of language models to describe the process of applying prompting to effectively extract information from language models, typically for use in real-world applications.
679
Meta's new LLM-based test generator
Meta's TestGen-LLM is a sneak peek to the future of developer productivity: specialized, orchestrated, and rigorously filtered.
680
A Remarkable Assertion from A16Z
A friend made me aware of a reading list from A16Z containing recommendations for books, weighted towards science fiction since that’s mostly what people there read.
681
Copilot Internals
Hacky repo to see what the Copilot extension sends to the server
682
My First Year as a Freelance AI Engineer
This week marks my one-year anniversary of quitting my full-time job and becoming a freelance AI engineer. I’m writing down my thoughts and experience here so that this might be useful if you are even vaguely interested.
683
Nvidia's Risky Business
Nvidia is finding new ways for its customers to raise money, and it’s expanding the risk of the AI buildout significantly.
684
Why deep learning works even though it shouldn’t
Why models always get better when they are bigger and deeper, even when the amount of data they consume stays the same or gets smaller.
685
Using machine learning to recreate photorealistic portraits of Roman Emperors
Artistic interpretations of the 54 Roman emperors of the Principate, created with AI and historical references.
686
Fine-tuning Mistral 7B on Magic the Gathering Draft
Tips, examples, and thoughts from an exploration of the world of fine tuning
687
A ChatGPT mistake cost us $10k
Edit: I want to preface this by saying yes the practices here are very bad and embarrassing (and we've since added robust unit/integration tests and alerting...
688
Two kinds of AI users are emerging
A bifurcation is happening in AI adoption - power users shipping products in days versus everyone else generating meeting agendas. Enterprise tool choices are accelerating the divide.
689
Apple Silicon costs more than OpenRouter
Local LLMs can be very very cheap
690
The neural network of the Stockfish chess engine
Last time, we briefly mentioned the high-level differences between Stockfish and Leela Chess. To recap, Stockfish evaluates about 100 million positions per second using rudimentary heuristics, wher…