Posts

Around 4 months ago, I thought “vibe coding” was a bad idea. Constant interactions with LLMs were not a good thing: we need more time and space to pause and think for ourselves. I have changed my mind. The tools have just got so much better. Perplexity voice is now a much better assistant on mundane tasks compared with Google. And the “vibe coding” experience has fundamentally changed too. It is no longer about copying and pasting from ChatGPT and hoping for the best: one can just stay in the IDE and monitor code changes step by step....

I remember coming across Scale of the Universe and being impressed by it. Is there something similar for financial information? If not, why not? There are so many similar and interesting questions: How does the wealth of Elon Musk compare to Li Ka-shing (or Sam Altman)? Which number is bigger, the market capitalisation of Apple or the annual tax takings of the US government? This is the kind of thing that is now hard to Google for, but ChatGPT and Perplexity come to the rescue:...

Last night, I changed a line of documentation in the Python source code. It is already reflected on the Python website, and will be included in the future whenever people download Python. This short essay explains why and how. 1 Back story It was around seven months ago. I had stopped programming for a while, in despair over the attention economy and the impermanence of digital content. Coming from a humanities background, I was first attracted to programming by its allure of certainty and universality....

There is something cyberpunk about learning about operating systems. Before I started, I assumed the materials would be very systematic and well laid out: start with how chips are designed, and then on to assembly language, and then on to compilers, and then finally how the operating system lives on top of that. But it is not like this at all. An implicit knowledge of chips, assembly language, compilers etc. are assumed....

When I started to programme in 2022, a common (and correct) advice was to “get your hands dirty”. Start by making simple things happen (doesn’t matter what). The only way to “get” programming was to try to make simple things work, understand (viscerally) how the tiniest typos or logic errors can derail the project, and take things from there. This is no doubt still true to some extent. Someone who hasn’t wrangled with the terminal and/or setting up an IDE arguably doesn’t “get” programming at all....

An operating system is how a computer works. So one might be tempted to speak of an “operating system” for a segment of the community (e.g. the legal world) or for an individual (e.g. what makes someone “tick”). But such metaphors are not just shallow, but misleading. Comparing astrology with statistics does an injustice to how poetic omens are, and also how logical, rigorous, counter-intuitive, and surprising statistics often is....

The paper Differentiable Logic Cellular Automata popped up on my Hacker News feed tonight. It immediately attracted my attention due to my interest in the Game of Life. The paper begins with an immediately captivating demo showing how the Google logo can be generated in a Game-of-Life fashion, i.e. by the use of simple rules that one would not ordinarily expect to lead to sophisticated patterns. The claim seems to be that, given (say) an image, neural networks can somehow learn what Game-of-Life style rules are necessary to generate such an image....

It speaks to China’s AI prowess that the first biography of Sam Altman with access to the subject is this Chinese-language text. The book is divided into three parts. Part 1 traces Altman’s journey as a Silicon Valley founder to the early years of OpenAI. What struck me the most was how closely Altman was integrated into the Silicon Valley ecosystem. Altman first made his name by founding a startup called Loopt, where he partnered with a technical co-founder to create a business to share their location with friends....

On a conceptual level, the three are entirely separate activities. Programming is giving computers instructions to achieve a certain task, line-by-line, within a particular language. Generally, a single typo or mistake would crash the programme. Computer science (“CS”) is the more abstract, often mathematical, study of various issues arising out of computers. E.g. what is computable; how much time or memory do algorithms take; how electric circuits or distributed systems work etc....

One project within contemporary AI is to enable everyone to programme with nothing other than English prose. DeepSeek comes close to achieving this goal. This is how far I got with roughly 2 hours of DeepSeek Coder. For background, I only have a basic grasp of the broad web development landscape, and only one prior project with the technology involved (Python Flask). My first 8 prompts were as follows:...