The latest blog posts to take off on Hacker News. 204 posts.
First, I’m a human typing this post. I’m not going to tell you who I am. Sure, many will say that is cowardly and fair, but I actually don’t think it would bring much value. What matters more is that I describe why, and what I did and didn’t do during the last two weeks or so with the OpenClaw agent I set up, known as…
Since AlphaZero, lc0-style chess engines have been trained with RL. Specifically, you have the engine (search + model) play itself a bunch of times, and train the model to predict the outcome of the game.
Motivation I’ve been captivated by the recent movement to popularize mathematics formalization through the Lean theorem prover, and this year I’m diving deeper into learning it. For those unfamiliar with this revolution, I highly recommend watching Kevin Buzzard’s talks on YouTube for an overview of why formal…
One bad thing about archeologists is that some of the successful ones get a big head. People used to think the Olmecs, who made these colossal stone heads, were contemporary with the Mayans. But in…
I discuss several commonly held perspectives on the complex numbers and explore how their differences engage with several aspects of structuralism in the philosophy of mathematics.
Does a new algorithm that improves on Dijkstra'a classic one actually deliver a practical benefit for networking?
Foundations: Games, Rationality, and Traps In organizations, strategies and execution are interdependent. Outcomes depend not just on what you do, but on what others do, and what they expect you to…
We usually learn that 1 kilobyte is 1024 bytes, 1 MB is 1024 kilobytes, etc. But is this true or a kilobyte is precisely 1000 bytes?
An interactive visualization to understand how neural networks work
This post is a rewrite of my earlier blog post from Nov 2023 with new insights and updates.
Base changing and other calculations
This is a very important technique: only fill cells when you’re certain; no guessing.
Many AI enthusiasts debate whether Large Language Models actually "reason." My research indicates that a reasoning process does indeed occur, but its goal is different than we assume.
Another interesting way to graphically derive the Golden Ratio using an equilateral triangle inscribed in a circle and the chord which passes through the midpoints of the sides of the equilateral triangle.
One of the questions of the 2026 acx prediction contest is whether Nvidia’s stock price will close below $100 on any day in 2026. At the time of writing, it trades at $184 and a bit, so going down to $100 would be a near halving of the stock value of the highest valued company in the world.
A floating point number f has the form f = m · 2 e where m is called the mantissa and e is a signed integer exponent . We like to read numbers scaled by powers of ten, not two, so computers need algorithms to convert binary floating-point to and from decimal text. My 2011 post “ Floating Point to Decimal Conversion is…
Slot machines on the Las Vegas Strip return about 93 cents on the dollar. This is widely considered some of the worst odds in gambling. Yet on Kalshi, a CFTC-regulated prediction market, traders have wagered vast sums on longshot contracts with historical returns as low as 43 cents on the dollar. Thousands of…
A high-performance and high-level purely functional data-parallel array programming language that can execute on the GPU and CPU.
Tauformer is a topological transformer (see paper ) that replaces dot‑product attention with a Laplacian-derived scalar (taumode) per token/head, then attends using distances in that scalar space. Below is a post-style overview of the idea and the first training signals from a 30M-parameter run.
In the Game of Life, which still lifes can be produced by crashing gliders together? We’ve known for a few years that the answer cannot be “all of them”, because in 2022 Ilkka Törmä and Ville Salo found a patch of still life that, if it exists in the universe, must have existed since the beginning of time. And so…
Summary : Inspired by the definition of temperature in statistical mechanics and the possibility for it to be below zero, we try sampling LLaMA at T = − 0.001 T=-0.001 T = − 0.001 . The results are maximally weird.
Made in Notion, the connected workspace where AI agents and teams organize docs, projects and knowledge together.
New : You can now watch a recording of my talk on my YouTube channel ! Or you can just click "play" below, I suppose.
At the core of the attention mechanism in LLMs are three matrices: Query, Key, and Value. These matrices are how transformers actually pay attention to different parts of the input. In this write-up, we will go through the construction of these matrices from the ground up.
