Zach Williams

🏀📊

[Zach Williams @zachwill]

"Perhaps life is lived on a logarithmic time scale, compressed toward the end." https://t.co/sToxfHn5Cm

[Zach Williams @zachwill]

@robertgraham Have you read through E4M and early TrueCrypt code? Lots of stylistic overlap with v0.1 Bitcoin.

- random.c file in E4M
- TCFormat.c file in TrueCrypt
- util.cpp file in v0.1 Bitcoin

[Zach Williams @zachwill]

@karpathy Here was my scatterplot with data from AutoCodeBench — which honestly wasn’t the set of languages I was expecting at all:

[dex @dexhorthy]

coding models + coding harness together were deeply RL’d on read/write/edit/glob/grep/bash so they could write better code

Throw a frontier model in there and it can do all kinds of non-code stuff related to general operations / white collar stuff*

On the other side of the fs-as-ai-brain, Almost any tool can consume files, humans can read/reason/organize

*some specialized stuff for specific work like excel is needed but also is straightforward to bundle for a bash tool

[David @DavidSHolz]

a mentor once told me that there are three kinds of projects in life: two days, two months, or ten years

[Bilawal Sidhu @bilawalsidhu]

Pretty cool hack to blend between different video feeds to give you the feeling of free viewpoint video AKA. god's eye view.

TL;DR 36 cameras deployed at basketball & badminton venues for China's National Games, letting viewers drag around on their phones for different angles with 0.1s switching.

They've also got a more post-processed version for the broadcast feed itself with much nicer interpolation - classic bullet time that works great for viral social clips and replays.

But the best part of the whole press release? This is all based on the "strategic foundation of China Media Group's "5G+4K/8K+AI" strategy" :-)

The tech for this has been around for a while and works at scale. But i'm curious - would you actually want to control your own camera angles during a game? Or would you rather have the director just show you the good stuff?

[VB Knives @Empty_America]

Tiramisu was invented in the 1960s.

No one on earth had ever drunk a Margarita or eaten a Fajita in 1930.

You can just create traditions.

[Valentin Ignatev @valigo]

Computer Science is not science, and it's not about computers. Got reminded about this gem from MIT the other day

[Zach Williams @zachwill]

https://t.co/Y5EpV5MZQT

[Zach Williams @zachwill]

Wrote up my notes about Datastar

[Stammy @Stammy]

what are we doing with this blue here

[Basil🧡 @LinkofSunshine]

I have not stopped thinking about this guy's accent for one second

[Oyindamola🙄 @dammiedammie35]

This is absolutely !nsane man 🤯🏓

[Andrej Karpathy @karpathy]

"Move 37" is the word-of-day - it's when an AI, trained via the trial-and-error process of reinforcement learning, discovers actions that are new, surprising, and secretly brilliant even to expert humans. It is a magical, just slightly unnerving, emergent phenomenon only achievable by large-scale reinforcement learning. You can't get there by expert imitation. It's when AlphaGo played move 37 in Game 2 against Lee Sedol, a weird move that was estimated to only have 1 in 10,000 chance to be played by a human, but one that was creative and brilliant in retrospect, leading to a win in that game.

We've seen Move 37 in a closed, game-like environment like Go, but with the latest crop of "thinking" LLM models (e.g. OpenAI-o1, DeepSeek-R1, Gemini 2.0 Flash Thinking), we are seeing the first very early glimmers of things like it in open world domains. The models discover, in the process of trying to solve many diverse math/code/etc. problems, strategies that resemble the internal monologue of humans, which are very hard (/impossible) to directly program into the models. I call these "cognitive strategies" - things like approaching a problem from different angles, trying out different ideas, finding analogies, backtracking, re-examining, etc. Weird as it sounds, it's plausible that LLMs can discover better ways of thinking, of solving problems, of connecting ideas across disciplines, and do so in a way we will find surprising, puzzling, but creative and brilliant in retrospect. It could get plenty weirder too - it's plausible (even likely, if it's done well) that the optimization invents its own language that is inscrutable to us, but that is more efficient or effective at problem solving. The weirdness of reinforcement learning is in principle unbounded.

I don't think we've seen equivalents of Move 37 yet. I don't know what it will look like. I think we're still quite early and that there is a lot of work ahead, both engineering and research. But the technology feels on track to find them.

https://t.co/JCxTdKpuzv

[Kelsey Piper @KelseyTuoc]

by far the most realistic part of ancient greek myths is the part where the prophets tell them exactly what's going to happen and they get really mad at them and ignore them and then it comes true and they get even madder and ignore them harder.

[Austen Allred @Austen]

“The ingredient list is roughly the same other than the artificial dyes and Butylated hydroxytoluene” is one of the funniest things ever written in the New York Times

[Dylan O'Sullivan @DylanoA4]

I just want to tweet out this C. S. Lewis quote every weekend

[Zach Williams @zachwill]

@dtvelectronics @Narodism I think @Narodism was really on to something.

1. Sassaman wrote very little Mixmaster code. Mostly minor doc tweaks.

2. Hal’s code is nothing like the Nov 2008 code.

3. The Common and Crypto files in TrueCrypt 1.0 are eerily similar.

I’m fairly certain it’s Le Roux. Crazy.

[Zach Williams @zachwill]

@Narodism @dtvelectronics Especially compared to the November 16, 2008 source code.

[Zach Williams @zachwill]

@Narodism @dtvelectronics TrueCrypt/Common/FORMAT.c from v1.0a has a lot of similarities: https://t.co/1L6MQtSL0i

[Zach Williams @zachwill]

@hamiltonulmer This is awesome! Thanks for sharing — I’d never read it before.

Your CodeMirror DuckDB interface is super cool, man. I’ve been exploring a similar “instant feedback” Soulver-like UI.

Great call on the CodeMirror merge extension. Hadn’t thought of using that with React.

[Zach Williams @zachwill]

"Perhaps life is lived on a logarithmic time scale, compressed toward the end." https://t.co/sToxfHn5Cm