People keep debating whether Large Language Models have an internal model of the world (encoded somewhere in their weights or their correlations) or not. Some go as far as to entertain the idea that a model may be sentient though, in my opinion, we’re still far from that. Today I was wondering a different thing: how good is a Large Language Model’s eye for design? I work as a full-stack web…
Since I started this site, I have endeavored to make it load as fast as possible. Call it a tradesman’s pride, or a small obsession. Recently, after a long period of not changing anything on the design or implementation, I decided to fix a few things I knew to be bottlenecks. None of these were major, I knew. Site speed was at ~1.2s according to webpagetest.org , but I also knew changes in loading…
Many forums or blogs make use of the spoiler tag: a little button or anchor that, if clicked, reveals otherwise invisible content. I wanted to add this functionality to the site for Tables of Content, so I figured adding this guide here could be useful both for my own future reference and for anyone else looking for a concise explanation. Here’s an example of what I mean by spoiler tag. Click to…
What could we do if, instead of coding programs from the ground up, we could just specify the rules for a task, the success criteria, and make AI learn to complete it? Imagine the blessings humanity could unlock if we could automate all the tasks nobody wants to do, all the unsafe, unhealthy or uninspiring jobs. Or the ways Scientific progress could be sped up if big parts of the research process…
The Japanese language has many quirks that make it alluring for students. After studying it for almost 8 years, one of the things that fascinates me is its rich pool of onomatopoeic expressions. English is not lacking in this respect: we have words like bang, clang, clatter and sputter which sound like the verbs they describe. Riffraff, whippersnapper and so on also have the property of…
Ant Colony Optimization algorithms always intrigued me. They are loosely based in biology and the real protocols ants use to communicate and plan routes. They do this by coordinating through small pheromone messages: chemical trails they leave as they move forward, signaling for other ants to follow them. Even though each ant is not especially smart, and they follow simple rules individually,…
I wrote this post when Stability AI had just released the AI art model Stable Diffusion. In this guide, you will learn how to write prompts by example. If you want, you can skip the discussion and examples and just go straight to the prompt template and explanation Stable Diffusion is similarly powerful to DALL-E 2 , but open source, and open to the public through Dream Studio , where anyone gets…
I have been using a personal wiki for note-taking and studying for 2 years. In that time, it has helped me a lot when preparing exams or job interviews, and generally made studying from books or papers feel more productive and enjoyable. A part of that is just emotional: it feels good to take notes and capture what you learn, knowing you will be able to find it again later. Being more deliberate…
OpenAI’s Dall-E 2 became available in beta a month ago, and as a big fan of generative text-to-image models I instantly joined the waitlist. After a little less than a month, the email arrived and I was notified I was finally allowed to use DALL-E 2! OpenAI gave me 50 free credits, with a monthly refill of 15 more from September onwards. I won’t go into how text-to-image models work here, and for…
I have been maintaining a Personal Wiki / Digital Garden for two years. Throughout all this time, I’ve reflected a lot on how I keep my notes, why I do it and what I get from doing it. Here is what I have learned. To save you time, I will start by stating my conclusion. It could be summarized as: Keeping notes is great, definitely do it. Maybe use a personal wiki? Before I started running this…
Last year I took an elective on Computer Graphics ( course notes ) where I learned about OpenGL shaders, and image compression algorithms. One of the algorithms I learned about was Floyd–Steinberg dithering . Sometimes when storing an image, we want to compress it lossily: we create a new version of it that loses a part of the information and quality, but is significantly smaller in filesize. One…
Suppose you have data, and want to report a certain metric. For instance, you’d like to estimate the value of some KPI given a set of measures, or you’re a physicist trying to estimate a constant from the readings of a sensor. Since you’re estimating that quantity using empiric data, sampled from the “true” distribution, you will have sample error in your estimation. This means even if your…
Ever since DALL-E came out in January 2021, or even before that with PixelRNN , I’ve found generative models, especially for images, amazing. But what I was interested in was how could I write better prompts that yielded the most beautiful images? The idea of text-to-image generators like CLIP or GLIDE is even more astounding, and I love being able to play with them and trying to get a glimpse of…
As I’ve said previously, this site is just a casual, fun thing I’m maintaining, mostly a repository for my thoughts and notes. That’s why the Digital Garden/ Zettelkassen / Personal wiki part gets most of the attention. With that said, today I finally sat down to improve this a bit. Here’s what I finally did today: Figured out how to run the site locally. My iteration speed and testing/debugging…
Honestly, this is just a fun experiment. I may use this as a laid back blog where I rant about my life, philosophy, Politics, programming and who knows what else. This website is 90% me playing around with Jekyll, 10% content, so don’t be surprised if this is the only blog post for a while. I may also use this site as a bit of a wiki/knowledge repository for the times when Evernote doesn’t seem…
There is a Japanese word, tsundoku (積ん読), which means buying and keeping a growing collection of books, even though you don’t really read them all. I think we Developers and Data Scientists are particularly prone to falling into this trap. Personally, I even hoard bookmarks: my phone’s Chrome browser has so many open tabs, the counter was replaced with a “:D” emoji. In that zeal for reading and…
Note: I originally wrote this article on my first blog, so it is not as polished as newer things. Python’s FuzzyWuzzy library is used for measuring the similarity between two strings. Here’s how you can start using it too. Sometimes, we need to see whether two strings are the same. When comparing an entered password’s hash to the one stored in your login database, ‘similarity’ just won’t cut it.…
Note: I originally wrote this article on my first blog, so it may not be as polished as newer things. To some, Vim is a beautiful relic from the past. To others, it’s that annoying thing you have to escape whenever you need to write a message for a merge commit. Let me introduce you to this picturesque text editor and its wonders, and show you why we’re still using it 26 years after its first…
Unsupervised Learning has been called the closest thing we have to “actual” Artificial Intelligence, in the sense of General AI, with K-Means Clustering one of its simplest, but most powerful applications. I am not here to discuss whether those claims are true or not. I will however state, that I am often amazed by how well unsupervised learning techniques, even the most rudimentary, capture…