Every blog post that took off on Hacker News, ranked by the votes of its best submission. 3,092 posts.
Like you, I'm sick to the back teeth of talking about AI. Like you, I keep getting dragged into discussions of AI. Unlike you‡, I spent the summer writing a book about why I'm sick of writing about AI⹋, which Farrar, Straus and Giroux will publish in 2026.
Update 2025: The post below describes ChatGPT and similar LLMs as they functioned at the time of writing in 2023. More recent systems do have a behind-the-scenes “inner monologue” which enables them t
The largely overlooked privacy risks of using AI apps that not only remember your conversations, but are capable of using these to reveal your deepest secrets to others
Some observations from my War of the Lancehttps://en.wikipedia.org/wiki/WaroftheLance%28videogame%29 1989 Code Harness assisted modernisation WIP I gave Qwen-3....
New from Google. They say it's "built to push the frontier of intelligence and solve modern challenges across science, research, and engineering". It drew me a really good SVG of …
I’ll let the horror of automation speak for itself. Not that you need further reassurance that there are no “souls” aboard this ship. WARNING: Some naughty language (this is compl…
I have a solid workflow for making my game with AI. However, every iteration, I have to go through extensive manual play-testing to verify functionality…
Part 1 of a multi-part series on using LLMs securely within your organisation. This post provides a framework to categorize risks based on different use cases and deployment type.
[Updated on 2023-01-24: add a small section on Distillation.] Large transformer models are mainstream nowadays, creating SoTA results for a variety of tasks. They are powerful but very expensive to train and use. The extremely high inference cost, in both time and memory, is a big bottleneck for adopting a powerful…
A from-scratch JEPA world model trained on Super Mario Bros. to study action-conditioned prediction and reward-free planning.
A language model on a 1975 MOS 6502 processor, using a tiny Mamba-based language model and custom 8-bit ternary inference engine, used to generate text on a BBC Micro from the 1980s.
Over the course of four weeks in January and February, I built a new programming language using Claude Code. I named it Cutlet after my cat. It’s completely legal to do that. You can find the source code on GitHub , along with build instructions and example programs .
An article from dylanbeattie.net
"Cognitive debt," Claude Code, and the negative space around AI
Google’s Nest acquisition has very little to do with selling thermostats and smoke detectors in particular. Instead, it’s about Google having the ability to do consumer hardware right, in general.
The first Times crossword Having seen several mind-blowing articles about ChatGPT’s ability (most notably this one), I wondered how ChatGPT would fare solving a cryptic crossword. I decided to try with an easy one: a Quick Cryptic from The Times. I had a search through the Times for the Times archives to try and find…
I've been working for a while on a program that automatically generates analog electronic circuits using evolutionary algorithms. It's still very much a work in progress, but I've used it to generate some useful circuits. Current workings of the program is as follows: Programmer decides a goal function, possible…
I had a really enlightening talk with Daniel Slutsky this week (who is an exceptional data scientist, software engineer, and community organizer I highly…
Reviewing the few changes in the Claude 4.0 system prompts, we get a sense for how system prompts program chatbot applications.
The lottery ticket hypothesis explains why massive neural networks succeed despite centuries of theory predicting they should fail
What if you could let Claude and Codex work together as pair programmers, talking to each other directly? One of them as the main worker and the other as a reviewer.
Fully discrete methods dominated for a few years, but language models based on continuous diffusion are making a comeback.
Why network effects and distribution will be king, once more
We explore language models that recursively call themselves or other LLMs before providing a final answer. The aim is essentially unbounded context, and a way through context rot.
Gradient descent is the preferred way to optimize neural networks and many other machine learning algorithms but is often used as a black box. This post explores how many of the most popular gradient-based optimization algorithms such as Momentum, Adagrad, and Adam actually work.
